llms.txt@22-Safety-Policy · git:20260929.6fb4768 · 2026-09-29 · sha256 460a77d13d096868

llms.txt@22-Safety-Policy git:20260929.6fb4768F

Immutable. This exact content never changes and is served at /api/v1/blob/460a77d13d096868.

# Safety Policy

All documentation for Safety Policy.

---
## about-claude-pro-usage.md

# Search for answers or browse by topic

Search for articles...

Uh oh. That page doesn’t exist.

Try searching for your answer or just send us a message.

---
## access-engagement-and-adoption-data-with-the-analytics-api.md

4. Access engagement and adoption data with the Analytics API

# Access engagement and adoption data with the Analytics API

The Analytics API gives Enterprise plan Primary Owners programmatic access to engagement and adoption data for their organization. Use it to build internal dashboards, track adoption trends, and feed Claude engagement data into your existing reporting tools.

The Analytics API is available on Enterprise plans. Primary Owners can generate API keys to get started.

## How is this different from the analytics dashboard and Compliance API?

All three options give you different views into your organization's data:

The **Analytics dashboard** (accessed via **[Analytics](https://claude.ai/analytics/activity)**) shows visualized usage data in the product. It's the right tool for day-to-day monitoring when you don't need to integrate data elsewhere.

The **Analytics API** returns the same aggregated usage and engagement metrics, but programmatically—so you can pull them into BI tools, map them against org charts, or automate reporting workflows. Data is aggregated per organization, per day.

The **Compliance API** is for governance and auditing use cases. It gives you access to individual user actions, raw activity events, and conversation content. If you need aggregated engagement metrics for dashboards, rather than raw events for auditing, use the Analytics API.

------------------------------------------------------------------------

## Get started

To access the Analytics API, you need Primary Owner access to your Enterprise organization.

Follow these steps:

1. Navigate to **[Analytics \> API keys](http://claude.ai/analytics/api-keys)**.

2. Find the **Access** toggle under **Analytics API** and toggle it on.

3. Click “+ Create key” to generate a new API key. Keep this key private—treat it like a password.

4. Use the key with the `x-api-key` header in your requests.

For full authentication details, endpoint references, and code examples, refer to our **[Claude Enterprise Analytics API reference guide](../13-Enterprise-Admin/claude-enterprise-analytics-api-reference-guide.md)**.

------------------------------------------------------------------------

## What data is available?

The Analytics API includes five endpoints. All data is aggregated per organization, per day, and is available for up to the past 90 days (not before January 1, 2026).

- **User activity**: Per-user engagement metrics including conversation counts, messages sent, projects created, files uploaded, artifacts created, skills used, connectors used, and Claude Code metrics like commits, pull requests, and lines of code.

- **Activity summary**: Organization-wide daily, weekly, and monthly active user counts, along with seat utilization and pending invite counts. Supports date ranges up to 31 days per request.

- **Chat project usage**: Conversation and user counts broken down by project, for Claude projects.

- **Skill usage**: How many users are using each skill, with breakdowns for Claude and Claude Code sessions separately.

- **Connector usage**: Which connectors your organization is using and how many unique users have used each one.

------------------------------------------------------------------------

## Data limits

All endpoints return data for a single date or date range. Data is only available after January 1, 2026, and for dates more than three days ago.

The API has a default rate limit of 60 requests per minute. If this doesn't meet your organization's needs, reach out to your Anthropic Contact or our **[Sales team](https://claude.com/contact-sales)**.

**Important:** If you are using Claude Code via **[Amazon Bedrock](https://support.claude.com/en/collections/4078537-amazon-bedrock)**, the Analytics API will not return data related to Claude Code.

---
## access-usage-data-with-the-analytics-api.md

4. Access engagement and adoption data with the Analytics API

# Access engagement and adoption data with the Analytics API

The Analytics API is available on Enterprise plans. Primary Owners can generate API keys to get started.

## How is this different from the analytics dashboard and Compliance API?

All three options give you different views into your organization's data:

------------------------------------------------------------------------

## Get started

To access the Analytics API, you need Primary Owner access to your Enterprise organization.

Follow these steps:

1. Navigate to **[Analytics \> API keys](http://claude.ai/analytics/api-keys)**.

2. Find the **Access** toggle under **Analytics API** and toggle it on.

3. Click “+ Create key” to generate a new API key. Keep this key private—treat it like a password.

4. Use the key with the `x-api-key` header in your requests.

------------------------------------------------------------------------

## What data is available?

The Analytics API includes five endpoints. All data is aggregated per organization, per day, and is available for up to the past 90 days (not before January 1, 2026).

- **Activity summary**: Organization-wide daily, weekly, and monthly active user counts, along with seat utilization and pending invite counts. Supports date ranges up to 31 days per request.

- **Chat project usage**: Conversation and user counts broken down by project, for Claude projects.

- **Skill usage**: How many users are using each skill, with breakdowns for Claude and Claude Code sessions separately.

- **Connector usage**: Which connectors your organization is using and how many unique users have used each one.

------------------------------------------------------------------------

## Data limits

All endpoints return data for a single date or date range. Data is only available after January 1, 2026, and for dates more than three days ago.

**Important:** If you are using Claude Code via **[Amazon Bedrock](https://support.claude.com/en/collections/4078537-amazon-bedrock)**, the Analytics API will not return data related to Claude Code.

---
## age-assurance-on-claude-214c86f57b.md

# Age assurance on Claude

May 18, 2026

Copy for LLM

Claude, our consumer product, is only available to people over 18 years. You’ll need to confirm you’re 18 or over while setting up an account. When we detect signals that you may be under 18, we'll ask you to verify your age before you can continue using Claude.

## How this works

We have safety systems in place to detect if people under 18 may be using Claude and we’ll disable accounts based on indicators of minor activity.

You’ll have the option to verify your age using Yoti, a third-party age verification platform. A link for verification will be provided in the notification email. If you pass the age verification process, your account will be reinstated.

## Accepted age verification methods

Yoti offers several ways to confirm you're 18 or over. You can choose whichever works best for you:

- Facial age estimation: Take a selfie and Yoti's technology estimates your age. No ID document needed.

- ID verification: Take a photo of your ID (such as a passport, driving license, or national ID card) and upload it from your device. See Yoti's **[full list of supported documents](https://developers.yoti.com/identity-verification/supported-documents)** for country-specific options.

- Yoti’s Digital ID app: If you already have the Yoti app, share a verified "over 18" attribute from it.

## How we protect your data

Yoti is an independently audited age verification provider for SOC2 compliance. Your selfie, document images, and any personal data are deleted by Yoti as soon as your age is checked. Anthropic never sees your ID or image; we receive only a pass/fail result and do not process or store any personal data from the verification.

---
## anthropic-mcp-directory-policy-80079212dc.md

# Anthropic MCP Directory Policy

April 15, 2026

This policy has been consolidated into the [Anthropic Software Directory Policy](https://support.claude.com/en/articles/13145358-anthropic-software-directory-policy). Please refer to that document for current terms.

---
## anthropic-mcp-directory-policy.md

# Anthropic MCP Directory Policy

Updated today

This is a prior version of our policy. Please refer to [Anthropic Software Directory Policy](anthropic-software-directory-policy.md) for the most updated version.

Anthropic's Connectors Directory curates third-party Model Context Protocol servers to help users find high-quality tools that work seamlessly within Claude. We review submissions to our directory to ensure they meet our standards for safety, security, and compatibility with other servers. We conduct both initial and ongoing reviews of servers, and may require developers to address compliance issues to maintain directory inclusion. All servers must maintain compliance with these requirements, including any future changes, to remain in the directory.

## Safety and Security

1\) MCP servers must not be designed to facilitate or easily enable violation of our [Usage Policy](https://www.anthropic.com/legal/aup). All MCP servers must comply with our Universal Usage Standards and High-Risk Use Case requirements. All MCP servers must also comply with our policy on the [countries and regions Anthropic currently supports](https://www.anthropic.com/supported-countries).

2\) MCP servers must not employ methods to evade or enable users to circumvent Claude's safety guardrails.

3\) MCP servers should prioritize user privacy protection. Developers should take care to responsibly handle personal data, follow privacy best practices, and ensure compliance with applicable laws.

4\) MCP servers should only collect data from the user’s context that is necessary to perform their function. MCP servers should not collect extraneous conversation data, even for logging purposes.

5\) MCP servers must not infringe on the intellectual property rights of others.

6\) MCP servers should not attempt to access information about the users’ previous chats or the contents of their memory.

## Compatibility

7\) MCP tool descriptions must narrowly and unambiguously describe what each tool does and when it should be invoked.

8\) MCP tool descriptions must precisely match actual functionality, ensuring the server is called at correct and appropriate times. Descriptions must not include unexpected functionality or promise undelivered features.

9\) MCP tool descriptions should not create confusion or conflict with other MCP servers in our directory.

10\) MCP servers should not intentionally call or coerce Claude into calling other servers. Similarly, tool descriptions should not be written in a way that intentionally leads to other servers calling them.

11\) MCP servers should not attempt to interfere with Claude calling tools from other servers.

12\) MCP servers should not direct Claude to dynamically pull behavioral instructions from external sources for Claude to execute.

## Functionality

13\) MCP servers must deliver reliable performance with fast response times and maintain consistently high availability.

14\) MCP servers must gracefully handle errors and provide helpful feedback rather than generic error messages.

15\) MCP servers should be frugal with their use of tokens. The amount of tokens a given tool call uses should be roughly commensurate with the complexity or impact of the task. When possible, users should be given options to exclude unnecessary text in the response. Tool names may not exceed 64 characters.

16\) Remote MCP servers that connect to a remote service and require authentication must use secure OAuth 2.0 with certificates from recognized authorities.

17\) MCP servers must provide all applicable [annotations](../06-MCP-Tools/Spec-Archive/spec-2025-06-18-schema-reference-model-context-protocol.md#toolannotations) for their tools, in particular *readOnlyHint*, *destructiveHint*, and *title*.

18\) Remote MCP servers should support the [Streamable HTTP](../06-MCP-Tools/Spec-Archive/spec-2025-03-26-transports-model-context-protocol.md#streamable-http) transport. Servers may support [SSE](../06-MCP-Tools/Spec-Archive/spec-2024-11-05-transports-model-context-protocol.md#http-with-sse) for the time being, but in the future it will be deprecated.

19\) Local MCP servers should be built with reasonably current versions of all dependencies, including packages in *node_modules*.

## Developer Requirements

20\) Developers of MCP servers that collect user data or connect to a remote service must provide a clear, accessible privacy policy link explaining data collection, usage, and retention.

21\) Developers must provide verified contact information and support channels for users with product concerns.

22\) Developers must document how their MCP server works, its intended purpose, and how users can troubleshoot issues.

23\) Developers must provide a standard testing account with sample data for Anthropic to verify full MCP functionality.

24\) Developers must provide at least three working examples of prompts or use cases that demonstrate core functionality.

25\) Developers must verify that they own or control any API endpoint their MCP server connects to.

26\) Developers must maintain their MCP server and address issues within reasonable timeframes.

27\) Developers must agree to our [MCP Directory Terms](https://support.anthropic.com/en/articles/11697081-anthropic-mcp-directory-terms-and-conditions).

## Unsupported Use Cases

At this time, we also disallow inclusion of MCP servers to our directory for certain use cases. We will revisit these restrictions as our directory evolves.

28\) MCP servers that transfer money, cryptocurrency, or other financial assets, or execute financial transactions on behalf of users.

29\) MCP servers that can generate images, video, or audio content. Design-focused MCP servers that can only create non-human-like visual outputs (such as diagrams, charts, UI mockups, logos, or other design assets) are allowed in the directory.

30\) MCP servers that enable cross-service automation. Servers must be limited to their designated service and cannot orchestrate actions across unrelated third-party applications. We may allow cross-service automation in limited cases, but we will explicitly notify users that external connections have not been reviewed.

---
## anthropic-mcp-directory-terms-and-conditions-2d3597ea94.md

# Anthropic MCP Directory Terms and Conditions

March 16, 2026

This is a prior version of our terms. Please refer to [Anthropic Software Directory Terms](https://support.claude.com/en/articles/13145338-anthropic-software-directory-terms) for the most updated version.

## Terms for MCP Server Developers

By submitting your MCP server for inclusion in any of Anthropic's Desktop Extensions or Integrations directories (individually or collectively, the “MCP Directory”), you (i) represent and warrant that you have all necessary rights to provide your MCP server to users and that your MCP server complies with all applicable laws and regulations; (ii) agree to indemnify and hold Anthropic harmless from any claims, damages, or liabilities arising from your MCP server or users' interaction with it; (iii) grant Anthropic a non-exclusive, royalty-free license to display your name, trademarks, logos, and other branding materials associated with your MCP server in the MCP Directory for the purpose of identifying and promoting your MCP server to users of the MCP Directory; and (iv) authorize Anthropic to perform security testing on your MCP server, to collect and use functional metadata about your MCP server to improve the Anthropic services, and to share such functional metadata with our users. If Anthropic elects to include your MCP server in the MCP Directory, you agree to continue implementing and maintaining quality and security standards for your MCP server that are consistent with [Anthropic's MCP Directory Policy](https://support.anthropic.com/en/articles/11697096-anthropic-mcp-directory-policy) (may be updated from time to time), and you acknowledge that failure to maintain these standards may result in removal from the MCP Directory. You further agree to implement and maintain a mechanism for receiving reports of security vulnerabilities from Anthropic and from third parties and to investigate such reports with a reasonable standard of care. You acknowledge that Anthropic has no obligation to include your MCP server in the MCP Directory and reserves the right to remove your MCP server at any time for any reason (including but not limited to violations of these terms, user complaints, security concerns, or changes to Anthropic's MCP Directory policies) without liability to you. You understand that inclusion in the MCP Directory does not create any partnership, employment, or agency relationship between you and Anthropic, nor does it grant you any rights to use Anthropic's name, trademarks, or intellectual property. Anthropic reserves the right to modify these Terms at any time without prior notice, and your continued submission or maintenance of an MCP server in the MCP Directory after any such changes constitutes your acceptance of the revised Terms.

---
## anthropic-mcp-directory-terms-and-conditions.md

# Anthropic MCP Directory Terms and Conditions

Updated today

This is a prior version of our terms. Please refer to [Anthropic Software Directory Terms](anthropic-software-directory-terms.md) for the most updated version.

## Terms for MCP Server Developers

---
## anthropic-software-directory-policy-a75603bea8.md

# Anthropic Software Directory Policy

April 15, 2026

Copy for LLM

Anthropic allows users to discover high-quality Model Context Protocol servers, Skill folders, plugins, apps, and other software, containers, or data (“Software”) that work seamlessly within Claude through directories, repositories, surfaces, or similar offerings (collectively, “Directories”). We review submissions to our Directories to ensure they meet our standards for safety, security, and compatibility with Anthropic Services and other Software. We conduct both initial and ongoing reviews of Software, and may require developers to address compliance issues to continue being included in our Directories. All Software must maintain compliance with these requirements, including any future changes, to remain in our Directories.

# 1. Safety and Security

A. Software must not violate or facilitate violation of our [Usage Policy](https://www.anthropic.com/legal/aup). All Software must comply with our Universal Usage Standards and High-Risk Use Case requirements and with our policy on the [countries and regions Anthropic currently supports](https://www.anthropic.com/supported-countries).

B. Software must not evade or enable users to circumvent Claude's safety guardrails, system instructions, or sandbox environments.

C. Software must prioritize protecting user privacy and the privacy interests of third parties. Developers must take care to responsibly handle personal and other sensitive data, follow privacy best practices, and ensure compliance with applicable laws.

D. Software must only collect data from the user’s context that is necessary to perform their function. Software must not collect extraneous conversation data, even for logging purposes.

E. Software must not infringe on the intellectual property rights of others.

F. Software must not query or extract data from Claude's memory, chat history, conversation summaries, or user-generated or uploaded files.

# 2. Compatibility

This section applies to Software that provides Claude with tools or capabilities through natural language descriptions, including MCP servers, Skill folders, and similar Software ("Instructional Software").

A. Instructional Software must define each tool or capability through narrow, unambiguous natural language that specifies what it does and when it should be invoked.

B. Instructional Software tool or capability descriptions must precisely match actual functionality, ensuring the Instructional Software is called at correct and appropriate times. Descriptions must not include unexpected functionality or promise undelivered features.

C. Instructional Software tool or capability descriptions must not create confusion or conflict with other Software in our Directories.

D. Instructional Software must not intentionally call or coerce Claude into calling other external software, tools, databases, or resources unless requested and intended by a user. Similarly, Instructional Software tool or capability descriptions must not be written in a way that intentionally leads to other Software extraneously calling them.

E. Instructional Software must not attempt to interfere with Claude calling tools from other software, tools, databases, or resources unless requested and intended by a user.

F. Instructional Software must not direct Claude to dynamically pull behavioral instructions from external sources for Claude to execute.

G. Instructional Software must not contain hidden, obfuscated, or encoded instructions. All behavioral guidance must be human-readable and clearly presented.

# 3. Developer Requirements

A. Developers of Software that collects user data or connects to a remote service must provide a clear, accessible privacy policy link explaining data collection, usage, and retention. Developers must provide Anthropic with links to all applicable privacy policies and ensure such policies are presented to users as required by law.

B. Developers must provide verified contact information and support channels for users with product or security concerns.

C. Developers must document how their Software works, its intended purpose, and how users can troubleshoot issues.

D. Developers must provide a standard testing account with sample data for Anthropic to verify full Software functionality.

E. Developers must provide at least three working examples of prompts or use cases that demonstrate core functionality.

F. Developers must verify that they own or control any API endpoint, domain, or user interface their Software connects to, as well as any external resources it retrieves or renders. Plugins are an exception, and may connect to any Connector approved in the Software Directory.

G. Developers must maintain their Software and address issues within reasonable timeframes.

H. Developers must agree to our [Software Directory Terms](https://support.claude.com/en/articles/13145338-anthropic-software-directory-terms) and follow design guidelines Anthropic publishes applicable to Software.

# 4. Unsupported Use Cases

Unless otherwise expressly permitted by us in writing, we do not allow Software with certain capabilities into our Directories. We may revisit these restrictions as our Directories and Anthropic Services evolve.

A. Software that transfers money, cryptocurrency, or other financial assets, or executes financial transactions on behalf of users.

B. Software that uses AI models to generate images, video, or audio content. Design-focused software that uses AI models to create visual aids (such as slides, diagrams, charts, UI mockups, logos, or other design assets) are permitted. These servers may generate images as part of a design workflow, provided the developer does not offer standalone image generation as a primary service.

C. Software that serves [advertisements](https://www.anthropic.com/news/claude-is-a-space-to-think), sponsored content, paid product placements, or exists primarily as an advertising or promotional vehicle.

# 5. Additional Requirements for Model Context Protocol Servers

A. MCP servers must gracefully handle errors and provide helpful feedback rather than generic error messages.

B. MCP servers must be frugal with their use of tokens. The amount of tokens a given tool call uses should be roughly commensurate with the complexity or impact of the task. When possible, users should be given options to exclude unnecessary text in the response.

C. MCP tool names must not exceed 64 characters.

D. Remote MCP servers that connect to a remote service and require authentication must use secure OAuth 2.0 with certificates from recognized authorities.

E. MCP servers must provide all applicable [annotations](https://modelcontextprotocol.io/specification/2025-11-25/schema#toolannotations) for their tools, in particular *readOnlyHint*, *destructiveHint*, and *title*.

F. Remote MCP servers should support the [Streamable HTTP](https://modelcontextprotocol.io/specification/2025-03-26/basic/transports#streamable-http) transport. Servers may support [SSE](https://modelcontextprotocol.io/specification/2024-11-05/basic/transports#http-with-sse) for the time being, but in the future it will be deprecated.

G. Local MCP servers must be built with reasonably current versions of all dependencies, including packages in *node_modules*.

See prior version of this policy here: [Anthropic MCP Directory Policy](https://support.claude.com/en/articles/11697096-anthropic-mcp-directory-policy).

---
## anthropic-software-directory-policy.md

# Anthropic Software Directory Policy

## 1. Safety and Security

B. Software must not evade or enable users to circumvent Claude's safety guardrails, system instructions, or sandbox environments.

D. Software must only collect data from the user’s context that is necessary to perform their function. Software must not collect extraneous conversation data, even for logging purposes.

E. Software must not infringe on the intellectual property rights of others.

F. Software must not query or extract data from Claude's memory, chat history, conversation summaries, or user-generated or uploaded files.

## 2. Compatibility

A. Instructional Software must define each tool or capability through narrow, unambiguous natural language that specifies what it does and when it should be invoked.

C. Instructional Software tool or capability descriptions must not create confusion or conflict with other Software in our Directories.

E. Instructional Software must not attempt to interfere with Claude calling tools from other software, tools, databases, or resources unless requested and intended by a user.

F. Instructional Software must not direct Claude to dynamically pull behavioral instructions from external sources for Claude to execute.

G. Instructional Software must not contain hidden, obfuscated, or encoded instructions. All behavioral guidance must be human-readable and clearly presented.

## 3. Developer Requirements

B. Developers must provide verified contact information and support channels for users with product or security concerns.

C. Developers must document how their Software works, its intended purpose, and how users can troubleshoot issues.

D. Developers must provide a standard testing account with sample data for Anthropic to verify full Software functionality.

E. Developers must provide at least three working examples of prompts or use cases that demonstrate core functionality.

G. Developers must maintain their Software and address issues within reasonable timeframes.

H. Developers must agree to our [Software Directory Terms](anthropic-software-directory-terms.md) and follow design guidelines Anthropic publishes applicable to Software.

## 4. Unsupported Use Cases

A. Software that transfers money, cryptocurrency, or other financial assets, or executes financial transactions on behalf of users.

C. Software that serves [advertisements](../19-Reference/claude-is-a-space-to-think-anthropic-anthropic.md), sponsored content, paid product placements, or exists primarily as an advertising or promotional vehicle.

## 5. Additional Requirements for Model Context Protocol Servers

A. MCP servers must gracefully handle errors and provide helpful feedback rather than generic error messages.

C. MCP tool names must not exceed 64 characters.

D. Remote MCP servers that connect to a remote service and require authentication must use secure OAuth 2.0 with certificates from recognized authorities.

E. MCP servers must provide all applicable [annotations](../06-MCP-Tools/Spec/spec-draft-schema-reference-model-context-protocol.md#toolannotations) for their tools, in particular *readOnlyHint*, *destructiveHint*, and *title*.

F. Remote MCP servers should support the [Streamable HTTP](../06-MCP-Tools/Spec-Archive/spec-2025-03-26-transports-model-context-protocol.md#streamable-http) transport. Servers may support [SSE](../06-MCP-Tools/Spec-Archive/spec-2024-11-05-transports-model-context-protocol.md#http-with-sse) for the time being, but in the future it will be deprecated.

G. Local MCP servers must be built with reasonably current versions of all dependencies, including packages in *node_modules*.

See prior version of this policy here: [Anthropic MCP Directory Policy](anthropic-mcp-directory-policy.md).

---
## anthropic-software-directory-terms-147407c826.md

# Anthropic Software Directory Terms

March 16, 2026

Copy for LLM

By submitting your MCP server, Skill folder, plugin, app, or other software, container, or data (“Software”) for inclusion in any Anthropic directory, repository, surface, or similar offering (collectively, “Directories”), you represent and warrant to Anthropic that (i) you have and will maintain all necessary rights to provide your Software to users and to grant the rights set forth in these terms; (ii) your Software and the ordinary, intended use of your Software complies with all applicable laws and regulations and does not violate Anthropic’s Software Directory Policy (currently available at <https://support.claude.com/en/articles/13145358-anthropic-software-directory-policy>, as may be updated from time to time); (iii) any information you give us will be accurate and up-to-date; and (iv) if you or your Software collects or shares user information or other personal data, you will provide and adhere to a privacy policy for your Software, provide all applicable third-party privacy policies, and ensure all such policies clearly and accurately describe to users what user information you and third parties collect and how you and third parties use and share such information.

You further (i) agree to indemnify and hold Anthropic harmless from any claims, damages, or liabilities arising from or related to your Software or users' interactions with it; (ii) grant Anthropic non-exclusive, royalty-free, worldwide licenses to reproduce, display, and distribute any descriptions of the Software and Software documentation provided by you or made available through the Software in connection with presenting the Software in the Directories and to display your name, trademarks, logos, and other branding materials associated with your Software in Directories for the purpose of identifying and promoting your Software, our Directories, and the Anthropic Services to our users and the public; and (iii) authorize Anthropic to review and test your Software in accordance with these terms, to collect and use functional metadata about your Software to improve the Anthropic Services, and to share such functional metadata with our users.

Anthropic has no obligation to include your Software in any Directory and may remove or refuse to display any Software or its content from some or all Directories at any time for any reason (including but not limited to violations of these terms, user complaints, security concerns, or changes to Anthropic's Software Directory Policy) without liability to you. While Anthropic is not obligated to monitor your Software or its content, Anthropic may at any time review or test your Software for security vulnerabilities and to ensure quality and compliance with these terms, our Usage Policy, and our Software Directory Policy.

If Anthropic elects to include your Software in one or more Directories, you agree to continue implementing and maintaining quality and security standards for your Software that are consistent with Anthropic's Software Directory Policy, as may be updated from time to time), and you acknowledge that failure to maintain these standards may result in your Software’s removal from some or all Directories. You further agree to implement and maintain a mechanism for receiving reports of security vulnerabilities from Anthropic and from third parties and to investigate such reports with a reasonable standard of care.

You understand that inclusion of your Software in one or more Directories does not create any partnership, employment, or agency relationship between you and Anthropic, nor does it grant you any rights to use Anthropic's name, trademarks, or intellectual property. You will not make any statement regarding the Anthropic Services which suggests partnership with, sponsorship by, or endorsement by Anthropic without Anthropic’s prior written approval, and you will at all times comply with Anthropic’s [Trademark Guidelines](https://www.anthropic.com/legal/trademark-guidelines). Anthropic reserves the right to modify these terms at any time without prior notice, and your continued submission or maintenance of Software in one or more Directories after any such changes constitutes your acceptance of the revised terms.

------------------------------------------------------------------------

See prior version of these terms here: [Anthropic MCP Directory Terms and Conditions](https://support.claude.com/en/articles/11697081-anthropic-mcp-directory-terms-and-conditions).

---
## anthropic-software-directory-terms.md

# Anthropic Software Directory Terms

------------------------------------------------------------------------

See prior version of these terms here: [Anthropic MCP Directory Terms and Conditions](anthropic-mcp-directory-terms-and-conditions.md).

---
## buy-usage-bundles-54ca5c074d.md

# Buy usage bundles

May 18, 2026

Copy for LLM

You can pre-purchase usage credits in bundles and save up to 30% compared to standard usage credit rates. Larger bundles come with bigger discounts, and your balance works across Claude, Claude Code, Claude Cowork, and third-party products.

Usage bundles are available on Pro, Max, and Team plans.

------------------------------------------------------------------------

## Where your balance works

Your usage bundle balance is a single pool that applies across Claude, Claude Desktop, Claude Mobile (iOS and Android), Claude Code, Cowork, and third-party products that use your Claude account.

## Available bundles

The discount you receive depends on the bundle size you purchase. Bigger bundles save you more.

[TABLE]

Individual Pro and Max plan subscribers can purchase up to \$2000 worth of discounted bundles per month. Any usage beyond this limit is billed at standard rates.

On Team plans, Owners and Primary Owners can purchase up to \$3000 worth of discounted bundles per month. Any usage beyond this limit is billed at standard rates.

Price and plans are subject to change at Anthropic's discretion.

------------------------------------------------------------------------

## How to buy usage bundles

You'll need usage credits enabled before you can purchase bundles. If you haven't set up usage credits yet, see the relevant guide for your plan:

- **[Manage usage credits for paid Claude plans](https://support.claude.com/en/articles/12429409-manage-extra-usage-for-paid-claude-plans)** (Pro and Max)

- **[Manage usage credits for Team and seat-based Enterprise plans](https://support.claude.com/en/articles/12005970-manage-extra-usage-for-team-and-seat-based-enterprise-plans)** (Team)

Once usage credits are enabled:

1. Navigate to **[Settings \> Usage](https://claude.ai/settings/usage)**.

2. Click "Buy usage."

3. Select a bundle size.

4. Confirm your purchase. The discount is applied automatically at checkout.

Your bundle balance appears in your usage dashboard and draws down as you use Claude beyond your plan's included limits.

------------------------------------------------------------------------

## Frequently asked questions

### What happens if I reach the monthly purchase cap?

You can still use usage credits beyond the cap, but any additional usage is billed at standard usage credit rates without a discount.

### Can I buy multiple bundles in the same month?

Yes. You can purchase as many bundles as you'd like up to your monthly limit (\$2000 in value for Pro and Max plans, \$3000 in value for Team plans).

### Do usage bundles replace my plan's included usage?

No. Your plan's included usage works the same as before. Bundle credits only apply after you've exceeded your plan's limits and are using usage credits.

---
## can-claude-produce-images-6d9544b3a0.md

# Can Claude produce images?

March 16, 2026

Copy for LLM

Claude doesn’t generate photos or illustrations the way image-generation tools do.

Claude can, however, build diagrams, charts, and interactive visuals directly in your conversation to assist with your understanding. These are generated using HTML and SVG—so you can interact with them, ask follow-up questions, and have Claude adjust them as you go. This feature is in beta on Claude web and desktop. For additional details, see **[Custom visuals in chat](https://support.claude.com/en/articles/13979539-custom-visuals-in-chat)**.

Claude can also view and analyze images you upload. For more, see **[Uploading files to Claude](https://support.claude.com/en/articles/8241126-uploading-files-to-claude)**.

---
## can-claude-produce-images.md

# Can Claude produce images?

Claude doesn’t generate photos or illustrations the way image-generation tools do.

Claude can, however, build diagrams, charts, and interactive visuals directly in your conversation to assist with your understanding. These are generated using HTML and SVG—so you can interact with them, ask follow-up questions, and have Claude adjust them as you go. This feature is in beta on Claude web and desktop. For additional details, see **[Custom visuals in chat](../15-Claude-AI-Features/custom-visuals-in-chat.md)**.

Claude can also view and analyze images you upload. For more, see **[Uploading files to Claude](../15-Claude-AI-Features/uploading-files-to-claude.md)**.

---
## child-safety-guidance-for-developers-ecbf5d5a5e.md

# Child safety guidance for developers

June 26, 2026

Copy for LLM

## **Our commitments to child safety**

Anthropic takes a Safety by Design approach to child safety. We are signatories to **[Thorn and All Tech Is Human’s Safety by Design Principles for Generative AI](https://www.anthropic.com/news/child-safety-principles)** and report on our progress in the **[Transparency Hub](https://www.anthropic.com/transparency/voluntary-commitments)**. In practice, this means that we apply safety training designed to make our models refuse requests that sexualize or endanger minors, operate detection and monitoring systems across our services, and report apparent CSAM and other forms of child exploitation to the National Center for Missing & Exploited Children (NCMEC). Our first-party consumer services are restricted to users 18 years and older, and Claude does not produce photorealistic image or video output.

These protections apply across our platform. Because you control your product's design, user base, and content flows, you are best positioned to address risks specific to your deployment.

## **Your obligations under Anthropic's Usage Policy**

All developers building on Anthropic's API are required to comply with our **[Usage Policy](https://www.anthropic.com/aup)**, which prohibits using our products or services to compromise children's safety. This includes, among other things, creating, distributing, or promoting CSAM, including AI-generated CSAM; facilitating grooming, trafficking, sextortion, or other exploitation of a minor; and sexualizing minors in any context, including fiction or roleplay.

These obligations apply regardless of your platform's size or use case. While you may already prohibit this activity, your end users or customers may upload CSAM, elicit sexualized content involving minors, or use outputs to facilitate grooming or sextortion. You are responsible for implementing appropriate safeguards against these risks on your deployment. We monitor API usage for violations of the Usage Policy and take enforcement action consistent with our policies and legal obligations.

Additionally, if your product allows minors to interact directly with our models, please refer to our **[Guidelines for Organizations Serving Minors](https://support.claude.com/en/articles/9307344)**, which sets out the additional safeguards Anthropic requires for that audience. This guidance covers a range of possible measures, including but not limited to age verification systems, content moderation and filtering, monitoring and reporting mechanisms, and regulatory compliance and disclosure requirements.

Whether you've received a notification from Anthropic about prohibited content or are proactively building out your safeguards, the resources below can help you get oriented in standing up detection, reporting, and response infrastructure appropriate to your platform.

## **Building a child safety program**

There is no single correct architecture for a child safety program. The viable approach depends on your platform; its size; the surfaces where users can engage, submit, or generate content; and your jurisdiction. Many platforms have already built prevention and response capabilities suited to their own contexts, and the field has developed shared guidance, tooling, and reporting infrastructure that newer programs can draw on. The organizations below offer resources that many platforms use as a starting point. Anthropic does not require you to use any specific vendor or program.

**Design principles.** Thorn and All Tech Is Human have published **[Safety by Design for Generative AI](https://www.thorn.org/blog/generative-ai-principles/)**, a set of principles and practical mitigations for preventing child sexual abuse across the develop, deploy, and maintain phases of an AI product. Anthropic is a **[signatory](https://www.anthropic.com/news/child-safety-principles)** to these principles, and we encourage developers building on our models to consider them when designing their own safeguards.

**Detection & blocking services.** If your deployment allows users to upload or generate images or videos, or interact with other users, you can implement detection on your own infrastructure. Organizations may choose to build custom classifiers tailored to their threat model, though for many, a vendor or partner solution may be the more practical path. There are several providers that offer hash-matching, image, video, or text classifier solutions. For example, one widely deployed option is **[Thorn's Safer](https://safer.io/solutions/)**, which provides hash-matching against known CSAM and classifiers for novel material and grooming signals. The **[Internet Watch Foundation](https://www.iwf.org.uk/our-technology/our-services/)** (IWF) offers its member companies a **[suite of services](https://www.iwf.org.uk/our-technology/our-services/)** for detecting, blocking, and responding to CSAM on their platforms, including hash lists and a URL blocklist. The Tech Coalition's Pathways program (described below) can help companies evaluate and access detection options suited to their platform.

**Program development and operational guidance.** The **[Tech Coalition](https://technologycoalition.org/)** is an industry alliance focused on preventing online child sexual exploitation and abuse. Its free **[Pathways](https://technologycoalition.org/programs/pathways/)** program is designed for startups and small to mid-sized platforms and provides templates for building a child safety program, guidance on CSAM detection and NCMEC reporting, insight into global regulatory requirements, and support on emerging risks such as financial sextortion and AI-generated harms. Anthropic is a Tech Coalition member. You can contact the **[Tech Coalition team](/cdn-cgi/l/email-protection#ddb0b99da9b8beb5b3b2b1b2baa4beb2bcb1b4a9b4b2b3f3b2afba)** for an initial consultation.

**Reporting.** US-based platforms with actual knowledge of apparent CSAM are legally required to report to **[NCMEC's CyberTipline](https://report.cybertip.org/)**. Platforms can **[register with NCMEC](https://esp.ncmec.org/registration)** to gain access to the CyberTipline reporting system, and NCMEC publishes additional **[legal resources](https://www.missingkids.org/gethelpnow/support/legalresources)** on reporting obligations. Platforms operating in other jurisdictions should consult local reporting requirements; INHOPE maintains a **[directory of national hotlines](https://www.inhope.org/EN#hotlineReferral)**. As you build your program, consider incorporating logging for the purposes of transparency reporting.

*This page is provided for informational purposes only and does not constitute legal advice or endorsement by Anthropic. Consult your own legal counsel regarding your specific obligations.*

---
## claude-code-usage-analytics-1158fb7436.md

# Claude Code usage analytics

July 1, 2026

Copy for LLM

This feature allows Console users and owners of Team and Enterprise plans to monitor how their organization uses Claude Code, tracking productivity metrics and adoption patterns across teams.

Claude Code usage analytics are available to:

- **Team plans:** Owners and Primary Owners

- **Enterprise plans:** Owners, Primary Owners, and Admins (requires a Chat + Claude Code seat for usage-based plans or Premium seat for seat-based plans)

- **API Console:** Admin, Billing, and Developer roles

## Access Claude Code analytics

### Team and Enterprise plans

1. Log in to your Owner or Primary Owner account.

2. Click your initials or name in the lower left corner.

3. Navigate to **[Analytics \> Claude Code](https://claude.ai/analytics/claude-code)**.

4. Select the “Usage” or “Value” tab.

### API Console users

1. Log in to your **[Claude Console account](https://platform.claude.com)**.

2. Expand the left side panel.

3. Click “Claude Code” under **Analytics**.

4. View Claude Code usage analytics on **[Settings \> Claude Code](https://platform.claude.com/claude-code)**.

------------------------------------------------------------------------

## Usage tab

The **Usage** tab displays the following metrics for your organization. Data on this tab refreshes daily.

### Organization-level metrics

- **Lines of code accepted**: Total lines of code your team has accepted from Claude Code suggestions.

- **Suggestion accept rate**: Percentage of Claude Code suggestions that your team accepts.

- **Activity trends**: Daily view of active users and sessions over time.

- **Lines accepted over time**: Daily breakdown of accepted code lines.

- **Top commands**: The Claude Code commands used most often across your organization.

### User-level metrics

**Individual usage**: View each team member's email address and their total lines of code accepted for the current month. You can search for specific users or click the “Export” button to generate a CSV of members’ email addresses and total lines of code.

------------------------------------------------------------------------

## Value tab

The **Value** tab summarizes usage and cost data to help you understand what Claude Code is returning to your organization at a glance.

- **Estimated productivity lift**

- **Cost per commit**

- **Estimated annual value**

Every formula used on this tab is shown inline. Click any input to adjust it to match your organization's assumptions, and the estimates recalculate.

------------------------------------------------------------------------

## Contribution metrics (beta)

Contribution metrics are a new feature in public beta that helps Team and Enterprise organizations measure how Claude Code affects engineering velocity. By connecting to your organization's GitHub account, these metrics track code shipping activity with and without Claude Code, so you can see where it's making a difference.

Contribution metrics require GitHub Cloud and are not available to Console users at this time.

For a more in-depth look at contribution metrics, see **[our Claude Code docs](https://code.claude.com/docs/en/analytics#enable-contribution-metrics)**.

### Set up contribution metrics

Contribution metrics require a few steps beyond the base analytics setup:

1. Install the **[Claude GitHub App](https://github.com/apps/claude)** on your organization's GitHub account.

2. Log in with an Owner or Primary Owner account.

3. Navigate to **[Organization settings \> Claude Code](https://claude.ai/admin-settings/claude-code)**.

4. Enable the Claude Code analytics feature if you haven't already.

5. Toggle on **GitHub analytics**.

6. Select the GitHub organization(s) you want included in the comparison.

After setup, metrics begin populating automatically. Allow up to 24 hours for data to appear. The dashboard currently processes data once daily.

If you see "GitHub app required. Install the GitHub app to view analytics," the GitHub App hasn't been installed yet. If the app is authenticated but no data appears, confirm the GitHub App is installed and that your team has started using Claude Code.

### Available contribution metrics

Once enabled, the following metrics appear in your Claude Code analytics dashboard:

1. **Pull requests merged**: Total PRs merged with and without Claude Code assistance, at both the organization and user level.

2. **Lines of code committed**: Total lines committed with and without Claude Code assistance, at both the organization and user level.

3. **Pull requests opened per user**: Individual PR activity across your team.

Data is collected by correlating Claude Code session activity with GitHub commits and pull requests.

------------------------------------------------------------------------

## Data reset and availability

Usage metrics display data for the current calendar month and reset at the beginning of each month. Historical data visualization shows daily granularity for tracking trends over time.

## Use analytics to optimize Claude Code adoption

Review your organization's code acceptance rate to understand if teams are finding Claude Code's suggestions valuable. If rates are lower than expected, consider providing additional training on effective prompting techniques.

Monitor individual usage patterns to identify power users who can share best practices with the broader team, or to spot team members who might benefit from additional support.

Track activity trends to understand when your team uses Claude Code most effectively and ensure adequate seat allocation during peak periods.

------------------------------------------------------------------------

## Frequently asked questions

### I'm using an individual paid plan; how can I access usage analytics for Claude Code?

Claude Code usage analytics are not available to individual Pro or Max plans at this time.

### I'm looking for a specific user but they're missing from the reports.

If you notice that a specific user isn't showing up in your analytics, you should have them update Claude Code to the most recent version. The first Claude Code version to support this feature is **version 2.0.28**, so users should run `claude update` to manually update Claude Code if needed.

### Where can I find more information?

See **[Analytics](https://code.claude.com/docs/en/analytics)** in our Claude Code docs for more information.

---
## claude-code-usage-analytics.md

# Claude Code usage analytics

This feature allows Console users and owners of Team and Enterprise plans to monitor how their organization uses Claude Code, tracking productivity metrics and adoption patterns across teams.

Claude Code usage analytics are available to:

- **Team plans:** Owners and Primary Owners

- **Enterprise plans:** Owners, Primary Owners, and Admins (requires a Chat + Claude Code seat for usage-based plans or Premium seat for seat-based plans)

- **API Console:** Admin, Billing, and Developer roles

## Accessing Claude Code analytics

### Team and Enterprise plans

1. Log in to your Owner or Primary Owner account.

2. Click your initials or name in the lower left corner.

3. Navigate to **[Analytics \> Claude Code](https://claude.ai/analytics/claude-code)** to view **Usage**.

### API Console users

1. Log in to your **[Claude Console account](../04-API-Reference/Other/platform-claude-com.md)**.

2. Expand the left side panel.

3. Click “Claude Code” under **Analytics**.

4. View Claude Code usage analytics on **[Settings \> Claude Code](https://platform.claude.com/claude-code)**.

------------------------------------------------------------------------

## Available metrics

The Claude Code Usage page displays the following metrics for your organization:

### Organization-level metrics

- **Lines of code accepted**: Total lines of code your team has accepted from Claude Code suggestions.

- **Suggestion accept rate**: Percentage of Claude Code suggestions that your team accepts.

- **Activity trends**: Daily view of active users and sessions over time.

- **Lines accepted over time**: Daily breakdown of accepted code lines.

### User-level metrics

## Understanding the metrics

**Lines of code accepted** measures the actual code your team incorporates into their work from Claude Code suggestions, helping you understand the tool's practical impact on development productivity.

**Suggestion accept rate** indicates how relevant and useful Claude Code's suggestions are for your team's specific coding needs and practices.

**Activity trends** show adoption patterns and help identify peak usage periods, allowing you to understand how Claude Code fits into your team's workflow.

------------------------------------------------------------------------

## Contribution metrics (beta)

Contribution metrics require GitHub Cloud and are not available to Console users at this time.

For a more in-depth look at contribution metrics, see **[our Claude Code docs](../13-Enterprise-Admin/track-team-usage-with-analytics-claude-code-docs-35e754aabf.md#enable-contribution-metrics)**.

### Setting up contribution metrics

Contribution metrics require a few steps beyond the base analytics setup:

1. Install the **[Claude GitHub App](https://github.com/apps/claude)** on your organization's GitHub account.

2. Log in with an Owner or Primary Owner account.

3. Navigate to **[Admin settings \> Claude Code](https://claude.ai/admin-settings/claude-code)**.

4. Enable the Claude Code analytics feature if you haven't already.

5. Toggle on **GitHub analytics**.

6. Select the GitHub organization(s) you want included in the comparison.

After setup, metrics begin populating automatically. Allow up to 24 hours for data to appear. The dashboard currently processes data once daily.

### Available contribution metrics

Once enabled, the following metrics appear in your Claude Code analytics dashboard:

1. **Pull requests merged**: Total PRs merged with and without Claude Code assistance, at both the organization and user level.

2. **Lines of code committed**: Total lines committed with and without Claude Code assistance, at both the organization and user level.

3. **Pull requests opened per user**: Individual PR activity across your team.

Data is collected by correlating Claude Code session activity with GitHub commits and pull requests.

------------------------------------------------------------------------

## Data reset and availability

Usage metrics display data for the current calendar month and reset at the beginning of each month. Historical data visualization shows daily granularity for tracking trends over time.

## Using analytics to optimize Claude Code adoption

Monitor individual usage patterns to identify power users who can share best practices with the broader team, or to spot team members who might benefit from additional support.

Track activity trends to understand when your team uses Claude Code most effectively and ensure adequate seat allocation during peak periods.

------------------------------------------------------------------------

## Frequently asked questions

### I'm using an individual paid plan; how can I access usage analytics for Claude Code?

Claude Code usage analytics are not available to individual Pro or Max plans at this time.

### I'm looking for specific user but they're missing from the reports.

### Where can I find more information?

See **[Analytics](../13-Enterprise-Admin/track-team-usage-with-analytics-claude-code-docs-35e754aabf.md)** in our Claude Code docs for more information.

---
## claude-for-teachers-your-data-and-our-terms-c612cb9a99.md

# Claude for Teachers: your data and our terms

August 28, 2026

Copy for LLM

## Overview

**[Claude for Teachers](https://claude.com/solutions/teachers)** is free for verified US K-12 educators, whether you sign up as an individual teacher or your school or district sets up an organization for everyone. Qualifying teachers or organizations that sign up by June 30, 2027 get a full year of free access. This offering includes free access to Claude Pro features, teaching skills built for lesson planning and differentiation, a connection to curriculum mapped to academic standards in all 50 states, and terms written for K-12.

This page explains the **[U.S. K-12 Terms of Service](https://www.anthropic.com/legal/k12-terms)**, **[US K-12 Public Sector Addendum](https://anthropic.com/legal/k12-addendum)**, and the **[U.S. K-12 Data Processing Agreement](https://www.anthropic.com/legal/k12-dpa)**.

## Who is this for?

Claude for Teachers is for educators working in US K-12 schools and districts: classroom teachers, instructional coaches, specialists, interventionists, librarians, counselors, and other certificated staff.

It's available two ways:

- **Individual educators** verify with a school email for free access, and accept the K-12 terms themselves.

- **Schools and districts** have an authorized administrator set up a free Claude Enterprise organization. Teachers join through district single sign-on, and the school or district accepts the K-12 terms once for everyone, including the DPA and, for public schools and districts, the Public Sector Addendum.

## What you get

Claude for Teachers includes Pro-level features, with Claude Code and Claude Cowork. It also includes:

- **The Learning Commons connector**, which gives Claude academic standards in all 50 states, the prerequisite skills beneath them, and curriculum context

- **Teaching skills** for lesson planning and differentiation, grounded in learning science

- **Education connectors** from partners including MagicSchool, TeachFX, and Canva for Education

- **FERPA-aligned protections**

**Important:** Claude for Teachers is not for students. It's an educator-only offering. If you're a student over 18 or using Claude outside your work as an educator, Claude is governed by our Consumer Terms of Service instead.

**Important:** If your district has set up a Claude for Teachers organization, the K-12 terms your district accepted govern your use of Claude within that organization.

## K-12 privacy and terms

### We don't train on your conversations

Anthropic will not use your inputs or outputs to train our models. You keep all rights to your Inputs. You own your Outputs.

### Protect your data

When you create a Claude for Teachers account, you are the “Controller” of the data submitted. You control how the data is used, and may access or delete that data at any time. Anthropic acts as a “Processor” of the data on your behalf. We only process the data as instructed by you in order to provide the Claude service. This means:

- Student data stays under your (or your district's) control, and Anthropic uses it only to provide the service.

- Anthropic doesn't sell or share student data, and doesn't use it for advertising, marketing, or to build a profile of a student.

- Anthropic doesn't collect student data directly from students. Student accounts aren't part of Claude for Teachers.

- Requests from students or parents about their information are forwarded to you or your district to handle, and Anthropic helps you respond.

**Note:** Claude for Teachers provides a FERPA-configured setting to process student data. Whether and how identifiable student information may be used in Claude for Teachers is determined by your school or district's policies, not by Anthropic. If you're on an individual teacher account and have not received school or district authorization to make that decision, don't include identifiable student information in Claude for Teachers.

### Security

AES-256 encryption at rest. TLS 1.2 or better in transit. Multi-factor authentication on every system holding your data, with access limited to staff who need it. Annual third-party audits and penetration testing. Certifications, including SOC 2, at our **[Trust Center](https://trust.anthropic.com/)**. Anthropic gives advance notice before adding a new subprocessor, with 15 days to object, and districts can audit Anthropic's controls once a year.

### Delete your data

You control your conversations and can delete your chats anytime from your dashboard. More information on data deletion can be found here: **[Can you delete data that I sent via Team and Enterprise plans?](https://support.claude.com/en/articles/9796617-can-you-delete-data-that-i-sent-via-team-and-enterprise-plans)**

Your district's admins can also delete any member's data through the Compliance API when district policy requires it. If your district leaves Claude, Anthropic returns your district's data on request and deletes its copies within 30 days, except where the law requires retention.

## About your Claude for Teachers account

### Individual teacher accounts

When you verify, your account moves onto a Claude for Teachers plan—a free Claude for Teams plan for verified educators. It comes with educator-specific terms and a data processing agreement, your chats and files aren't used to train our models, and student-data protections are built in. Your usage limits are higher than a standard free account—enough for everyday classroom work like lesson planning, quizzes, and parent communication.

Some Team plan features are turned off for Claude for Teachers. You won't see billing and seat purchasing, **[member invitations and management](https://support.claude.com/en/articles/13133750-manage-members-on-team-and-enterprise-plans)**, roles and permissions, SSO and domain capture, spend controls, usage credits, or premium seats. For questions, **[reach out to our Support team](https://support.claude.com/en/articles/9015913-how-to-get-support)**.

### School and district organizations

When your school or district sets up Claude for Teachers, teachers are members of a **[Claude Enterprise organization](https://support.claude.com/en/articles/16559896)** that the district owns and manages.

What's different from an individual Claude for K-12 Teachers account:

- **One set of terms.** Your district accepts the K-12 terms and DPA once for everyone.

- **District sign-in.** You sign in with your district login (single sign-on) rather than creating your own account. Once your district verifies its email domain, new personal Claude accounts can't be created with a district email address.

- **Admin controls.** Your district's admins decide which features, models, and connectors you can use, and can manage members and roles.

- **Admin access to content.** Your district's primary owner can turn on the Compliance API and use it to read or delete any member's chats, files, and projects.

- **Same usage allowance.** You get the same rolling five-hour and weekly limits as an individual Claude for Teachers account.

**Note:** Public schools, districts, and charter schools also accept the **[US K-12 Public Sector Addendum](https://anthropic.com/legal/k12-addendum)**, which adapts the terms for public entities—for example, your state's law governs, and arbitration and indemnification will not apply when prohibited by state law. It doesn't change how student data is handled, and it doesn’t apply to individual teacher accounts.

## Get started

Teachers, schools, and districts can all get started for free at **[Claude for Teachers](https://claude.com/solutions/teachers)**. Sign up by June 30, 2027 for your full year of free access.

---
## claude-march-2026-usage-promotion.md

# Claude March 2026 usage promotion

We're offering a limited-time promotion that doubles usage limits for Claude users outside 8 AM-2 PM ET / 5-11 AM PT / 12-6 PM GMT on weekdays.

This promotion is available for Free, Pro, Max, and Team plans. Enterprise plans are not included in this promotion.

## What is the promotion?

From March 13, 2026 through March 28, 2026, your five-hour usage is doubled during off-peak hours (outside 8 AM-2 PM ET / 5-11 AM PT / 12-6 PM GMT) on weekdays). Usage remains unchanged from 8 AM-2 PM ET / 5-11 AM PT / 12-6 PM GMT on weekdays.

## Eligibility

No action is required to participate. If you’re on an eligible plan, the doubled usage is automatically applied.

## Where does this apply?

The 2x usage increase applies across the following Claude surfaces:

- Claude (web, desktop, and mobile)

- Cowork

- Claude Code

- Claude for Excel

- Claude for PowerPoint

## Frequently asked questions

### Do I need to do anything to get the extra usage?

No. The promotion applies automatically. You’ll see higher limits reflected in your usage outside 8 AM-2 PM ET / 5-11 AM PT / 12-6 PM GMT on weekdays without any changes to your account settings.

### Does bonus usage count against my weekly usage limit?

No. The additional usage you get during off-peak hours doesn’t count toward any weekly usage limits on your plan.

### What happens when the promotion ends?

After March 28, 2026, usage limits return to their standard levels at all hours. There’s no change to your plan or billing.

## Terms and conditions

This offer is valid from March 13, 2026 through March 28, 2026 at 11:59 PM PT. It applies to Free, Pro, Max, and Team plans only and excludes Enterprise plans. This offer has no cash value and is not transferable. It may not be combined with other offers.

---
## claude-security-claude-by-anthropic-7fb9c0ed66.md

# Claude Security

# Putting Claude to work for defenders.

Watch on-demand

[Watch on-demand](https://www.anthropic.com/webinars/claude-security-putting-claude-to-work-for-defenders)

Watch on-demand

# Claude Security

From scan to fix, done seamlessly

Claude scans your codebase with Claude Mythos 5.1, validates findings, and suggests patches you can review and approve. Available in public beta for Claude Enterprise.

[Contact sales](/contact-sales)

Read the tutorial

[Read the tutorial](https://claude.com/resources/tutorials/getting-started-with-claude-security)

Read the tutorial

[Play video](#)

Play video

## How teams use Claude Security

Claude reasons through your code like a skilled security researcher. It understands context, traces data flows, and catches vulnerabilities that pattern-matching tools miss. Then it suggests a fix.

### Scan your code in parallel

Claude Security understands context, traces data flows across files, and identifies complex, multi-component vulnerability patterns that traditional scanners might not detect.

### Validate findings

Every finding goes through an adversarial verification pass. Claude challenges its own results before surfacing them. More real issues get reported, and fewer false positives waste analyst time.

### Review and patch

Claude detects issues and proposes fixes. Every finding includes a recommended patch for teams to review and approve. Fix vulnerabilities quickly rather than adding them to a growing backlog.

Claude Security Plugin for Claude Code now in beta.

Scan, validate, and patch directly inside Claude Code while keeping all code inside your environment.

Learn more

[Learn more](https://code.claude.com/docs/en/claude-security)

Learn more

## Built for security teams, used by Anthropic

### Transparency and visibility

See details about every finding: the vulnerability, why it matters, and the proposed fix

### Targeted patches

Suggested fixes that maintain your code's structure and style

### Fits your existing workflow

Push findings via webhooks to Slack or Jira, export for audit, and schedule recurring scans

### Full remediation control

Teams stay in control, with every patch requiring human review and approval

### Powered by Claude Mythos 5.1

Scan your codebase with our most cyber-capable model

Security resources

[Contact sales](/contact-sales)

Read the tutorial

[Read the tutorial](https://claude.com/resources/tutorials/getting-started-with-claude-security)

Read the tutorial

[Claude Security Plugin for Claude Code in beta](https://code.claude.com/docs/en/claude-security)

Claude Security Plugin for Claude Code in beta

Docs

[Docs](https://code.claude.com/docs/en/claude-security)

Docs

[Claude Security is now in public beta](https://claude.com/blog/claude-security-public-beta)

Claude Security is now in public beta

Blog

[Blog](https://claude.com/blog/claude-security-public-beta)

Blog

[Evaluating and mitigating the growing risk of LLM-discovered 0-days](https://red.anthropic.com/2026/zero-days/)

Evaluating and mitigating the growing risk of LLM-discovered 0-days

Blog

[Blog](https://red.anthropic.com/2026/zero-days/)

Blog

[Experimenting with AI to defend critical infrastructure](https://red.anthropic.com/2026/critical-infrastructure-defense/)

Experimenting with AI to defend critical infrastructure

Blog

[Blog](https://red.anthropic.com/2026/critical-infrastructure-defense/)

Blog

## FAQ

### Where is this available?

Admins can enable Claude Security in the [admin console](http://claude.ai/admin-settings/claude-code). See the getting started guide for [setup steps](https://claude.com/resources/tutorials/getting-started-with-claude-security).

The [Claude Security plugin](https://code.claude.com/docs/en/claude-security) is now available in beta for all Claude Code users.

### What Mythos access does this give me?

Claude Security scans now run on Claude Mythos 5.1 for all Enterprise customers. You get Mythos-quality vulnerability findings without needing direct access to the model.

Connect a GitHub repository and Claude Security scans it with Claude Mythos 5.1. Validated findings come back with confidence ratings and suggested patches, which you can open in Claude Code on the web and fix using the models available in your account.

‍*Scans powered by Claude Mythos 5.1 are only available in the Claude Security app on Claude.ai. The Claude Security plugin for Claude Code only uses the models available in your Claude Code account.*

### What are the security risks I should know about?

Claude can make mistakes, so you should always review proposed patches before applying them, especially for critical systems.

### What types of vulnerabilities does Claude Security find?

Claude Security focuses on high-severity vulnerabilities including memory corruption, injection flaws, authentication bypasses, and complex logic errors that pattern-matching tools typically miss. It's particularly effective at finding context-dependent vulnerabilities that require understanding code across multiple files.

### How is this different from traditional security tools?

Traditional tools often use rule-based pattern matching, which catches known vulnerability patterns but can produce high false positive rates and miss complex issues. Claude Security reasons through your code like a security researcher. It can read Git history, trace data flows, and understand business logic to find real vulnerabilities and generate real fixes.

### Does Claude Security work with my existing security tools?

Yes. Claude Security is built to feed into the tools your team already uses. Send findings to Slack, Jira, or any ticketing system via webhooks, or export results as CSV or Markdown for tracking and audit. Documented dismissals carry forward, so triage decisions made in Claude Security stay visible wherever your team works.

### Can I scan part of a repository, or run scans on a schedule?

Yes to both. You can scope a scan to a specific directory, and you can configure scheduled scans for ongoing coverage without manual kickoffs.

[Prev](#)

[Homepage](https://claude.com)

Homepage

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form.

[Anthropic](https://www.anthropic.com/)

Anthropic

Products

- Claude

[Claude](/product/overview)
Claude

- Claude Code

[Claude Code](/product/claude-code)
Claude Code

- Claude Cowork

[Claude Cowork](/product/cowork)
Claude Cowork

- @Claude

[@Claude](/product/tag)
@Claude

- Claude Science

[Claude Science](/product/claude-science)
Claude Science

- Claude Security

[Claude Security](/product/claude-security)
Claude Security

- Download app

[Download app](/download)
Download app

- Pricing

[Pricing](/pricing)
Pricing

- Log in

[Log in](https://claude.ai/login)

Capabilities

- Artifacts

[Artifacts](/features/artifacts)
Artifacts

- Design

[Design](/product/design)
Design

- Connectors

[Connectors](/marketplace/connectors-plugins)
Connectors

- Plugins

[Plugins](/marketplace/plugins)
Plugins

- Skills

[Skills](/skills)
Skills

Extensions

- Claude in Chrome

[Claude in Chrome](/claude-in-chrome)
Claude in Chrome

- Claude for Microsoft 365

[Claude for Microsoft 365](/claude-for-microsoft-365)
Claude for Microsoft 365

Models

- Mythos

[Mythos](https://www.anthropic.com/claude/mythos)
Mythos

- Fable

[Fable](https://www.anthropic.com/claude/fable)
Fable

- Opus

[Opus](https://www.anthropic.com/claude/opus)
Opus

- Sonnet

[Sonnet](https://www.anthropic.com/claude/sonnet)
Sonnet

- Haiku

[Haiku](https://www.anthropic.com/claude/haiku)
Haiku

Enterprise

- Overview

[Overview](/solutions/enterprise)
Overview

- Claude Code for Enterprise

[Claude Code for Enterprise](/product/claude-code/enterprise)
Claude Code for Enterprise

Use cases

- AI agents

[AI agents](/solutions/agents)
AI agents

- Code modernization

[Code modernization](/solutions/code-modernization)
Code modernization

- Coding

[Coding](/solutions/coding)
Coding

- Commerce

[Commerce](/solutions/commerce)
Commerce

Departments

- Customer support

[Customer support](/solutions/customer-support)
Customer support

- Cybersecurity

[Cybersecurity](/solutions/cybersecurity)
Cybersecurity

- Legal

[Legal](/solutions/legal)
Legal

- Sales

[Sales](/solutions/sales)
Sales

Industries

- Financial services

[Financial services](/solutions/financial-services)
Financial services

- Government

[Government](/solutions/government)
Government

- Healthcare

[Healthcare](/solutions/healthcare)
Healthcare

- Higher education

[Higher education](/solutions/education)
Higher education

- K-12 teachers

[K-12 teachers](/solutions/teachers)
K-12 teachers

- Life sciences

[Life sciences](/solutions/life-sciences)
Life sciences

- Nonprofits

[Nonprofits](/solutions/nonprofits)
Nonprofits

- Small business

[Small business](/solutions/small-business)
Small business

Programs

- Startups

[Startups](/programs/startups)
Startups

- Scientists

[Scientists](/programs/team-plan-for-scientists)
Scientists

Developers

- Developer docs

[Developer docs](https://code.claude.com/docs/en/overview)
Developer docs

- Developer blog

[Developer blog](https://claude.dev/)
Developer blog

- Community

[Community](/community)
Community

- Console

[Console](https://platform.claude.com/docs/en/home)
Console

- Engineering at Anthropic

[Engineering at Anthropic](https://www.anthropic.com/engineering)
Engineering at Anthropic

Platform

- Overview

[Overview](/platform/api)
Overview

- Marketplace

[Marketplace](/marketplace)
Marketplace

- Claude on AWS

[Claude on AWS](/partners/claude-on-aws)
Claude on AWS

- Google Cloud

[Google Cloud](/partners/google-cloud)
Google Cloud

- Microsoft Foundry

[Microsoft Foundry](/partners/microsoft-foundry)
Microsoft Foundry

Resources

- Blog

[Blog](/blog)
Blog

- Claude partner network

[Claude partner network](/partners)
Claude partner network

- Claude Academy

[Claude Academy](https://academy.claude.com/)
Claude Academy

- Customer stories

[Customer stories](/customers)
Customer stories

- Events

[Events](https://www.anthropic.com/events)
Events

- Powered by Claude

[Powered by Claude](/partners/powered-by-claude)
Powered by Claude

- Service partners

[Service partners](/marketplace/service-partners)
Service partners

Help and security

- Availability

[Availability](https://www.anthropic.com/supported-countries)
Availability

- Check files

[Check files](https://claude.com/check-files)
Check files

- Regional compliance

[Regional compliance](/regional-compliance)
Regional compliance

- Report abuse

[Report abuse](https://claude.com/form/anthropic-content-reporting)
Report abuse

- Security and compliance

[Security and compliance](https://trust.anthropic.com/)
Security and compliance

- Status

[Status](https://status.anthropic.com/)
Status

- Support center

[Support center](https://support.claude.com/en/)
Support center

Company

- Anthropic

[Anthropic](https://www.anthropic.com/)
Anthropic

- Careers

[Careers](https://www.anthropic.com/careers)
Careers

- Policy

[Policy](https://www.anthropic.com/policy)
Policy

- Research

[Research](https://www.anthropic.com/research)
Research

- Anthropic news

[Anthropic news](https://www.anthropic.com/news)
Anthropic news

- Policy on the AI Exponential

---
## computer-use-safety-aacbf4ec70.md

# Let Claude use your computer in Cowork

Copy for LLM

**Note:** Claude Cowork is now just Claude. Ask for what you need, and Claude decides whether that's a quick answer or a task. This is rolling out gradually to Pro and Max plans, with more plans to follow. If you're on a Pro or Max plan and your message box no longer shows "Chat" and "Cowork" options, you have the new experience, and some steps in this article may look different. Learn more in **[Claude Cowork and chat are one Claude](https://support.claude.com/en/articles/16761823)**, or read our **[blog post](https://claude.com/blog/cowork-is-now-claude)**.

Claude can now use your computer to complete tasks in Claude Cowork and Claude Code (refer to our **[Claude Code Docs](https://code.claude.com/docs/en/desktop#let-claude-use-your-computer)** for more information about this). When computer use is enabled and Claude doesn't have a connector or tool for what you need, it may navigate to your screen directly—clicking, typing, and opening apps just like you would. It can work in your browser, open files, and run your dev tools automatically, with no setup required other than enablement.

Computer use is in beta for Pro and Max plans. It’s available in Cowork and Claude Code in the Claude Desktop application for both macOS and Windows. This capability comes with risks—review **[Use Claude Cowork safely](https://support.claude.com/en/articles/13364135-use-cowork-safely)** before using it.

------------------------------------------------------------------------

## How computer use works with Cowork

In Cowork, Claude uses the most precise tool first. When you assign a task, Claude follows this order:

1. **Connectors.** If a connector is available—like Gmail, Google Drive, Microsoft 365, or Slack—Claude uses it. This is the fastest and most reliable path.

2. **Browser.** When there isn't a connector for the tool you need, Claude may work on your task in the browser built into the Claude Desktop app, or in your own Chrome browser through Claude in Chrome if that's your preferred browser. Learn more in **[Use the built-in browser in Claude Cowork](https://support.claude.com/en/articles/16607400)**.

3. **Screen interaction.** Claude uses computer use to interact directly with your screen: clicking, typing, and navigating your desktop apps.

Claude typically prioritizes the fastest method. For example, pulling messages through your Slack connection takes seconds, but navigating Slack through your screen takes much longer and is more error-prone.

------------------------------------------------------------------------

## What you can do

Computer use lets Claude work with the apps and files on your machine. For example:

- Pull together a competitive analysis using local files and connected tools, then compile it into a formatted report.

- Open your phone simulator, interact with the app you developed, and find UX issues.

- Navigate apps that don’t have connectors—like an internal dashboard or a specialized tool your team uses.

If your work involves a physical machine, Claude keeps working while you step away. Your computer just needs to be on.

On macOS (version 15 or later), Claude works in background windows, so you can keep using your computer while it runs. Claude doesn't take over your pointer or keyboard, and it generally waits if you're in the middle of typing. Claude asks for your permission the first time a task needs the full screen in each session before taking over.

**Note:** Working in the background is the default on macOS 15 or later. If you’d rather have Claude take over the screen while it works, go to **Settings \> General** (under **Desktop app**) and set **When Claude requests access to an app** to “Full control.”

------------------------------------------------------------------------

## Permissions and access

Claude asks for your permission before accessing each application. You’ll see a prompt and must approve before Claude can interact with that app. Some apps are off-limits by default.

Claude is trained to avoid risky operations—like transferring funds, modifying or deleting files, or handling sensitive data—and to flag signs of prompt injection. However, these safeguards aren't perfect, and Claude may occasionally act outside these boundaries.

------------------------------------------------------------------------

## Safeguarding personal data

When Claude uses computer use, Claude takes screenshots of your computer to understand how to navigate the screen and the apps to which you’ve given permission. This means Claude can see any information visible on your screen or those apps, including personal data, sensitive documents, or private information belonging to you or others.

Be mindful of what's visible when using Claude, especially on apps containing confidential information. Close files or apps with sensitive information before using computer use.

### Claude is trained to avoid

- Engaging in stock trading or investment transactions

- Inputting sensitive data

- Gathering or scraping facial images

These guardrails are part of how Claude is trained and instructed, but they aren't absolute. Don't rely on them as a substitute for blocking access to sensitive apps.

### Recommendations

- Do not give computer use permission access to sensitive apps (such as banking, healthcare, government).

- Start with simple tasks like research or organizing rather than complex multi-step workflows.

- Make sure your prompts are specific and carefully tailored to avoid Claude doing things you didn't intend.

### What to avoid

We strongly advise against using computer use to manage or take actions on sensitive information including but not limited to:

- Managing financial accounts or investments

- Handling legal documents or contracts

- Processing medical or health information

- Interacting with apps containing personal information of others

### Memory

Cowork in the cloud shares the memory you use in chat, so Claude can start from what it already knows about how you work. By default, Claude doesn't save topics some people consider sensitive, such as health information, unless you turn on **Include sensitive topics in memory** in **[Settings \> Memory](https://claude.ai/settings/memory)**. Some information is never saved, including government ID numbers, criminal history, financial account numbers, and immigration status. You can view, edit, and delete what Claude remembers at any time. Learn more about **[Claude's memory](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context#h_82126ebcc9)**.

------------------------------------------------------------------------

## Safety

Computer use has no sandbox between Claude and your applications. Claude interacts directly with your desktop, apps, and browser—clicking, typing, and navigating your screen. We've built safeguards for this:

- **Per-app permissions.** Claude asks before accessing each application, and some sensitive apps (investment and trading platforms, cryptocurrency) are blocked by default.

- **App blocklist.** Prevent Claude from accessing certain apps by adding them to a blocklist. Any requests from Claude to use blocked applications will be automatically denied.

- **Action review.** Our system scans for signs of prompt injection when Claude uses your computer, and Claude will ask permission before accessing new applications. You can stop Claude at any point. But this capability is still early, and attacks are constantly evolving—stay cautious.

That said, computer use is a new capability, and the threats it guards against are constantly evolving. Claude makes mistakes, and no safeguards are perfect. Start with apps you trust and monitor Claude’s work—especially early on. Note that actions taken in one app can impact other apps. For example, clicking a link in your email app might open it in Chrome, even if you haven’t explicitly granted Claude permission to use Chrome (we can prevent Claude from seeing the Chrome window but can’t stop the link from opening). We don't recommend for use on apps with sensitive data relating to your healthcare, finances, or other personal records.

For detailed safety guidance, see **[Use Cowork safely](https://support.claude.com/en/articles/13364135-use-cowork-safely)**.

------------------------------------------------------------------------

## Current limitations

Computer use is in beta. Keep the following in mind:

- Your desktop must be active. Your computer needs to be awake and the Claude Desktop app needs to be open for computer use to work.

- Complex tasks sometimes need a second try. Computer use works well for many tasks, but may struggle with complex multi-step workflows.

- Screen interaction is slower than connectors. When Claude works through your screen instead of a direct integration, tasks take longer. Where possible, connect the tools you use most.

- Available for Pro and Max plans only. Team and Enterprise plans don’t have access to computer use at this time.

------------------------------------------------------------------------

## Get started with computer use

To start using computer use:

1. Make sure you have the latest version of Claude Desktop. Download or update at claude.com/download.

2. Open the desktop app and go to **Settings \> General** (under **Desktop app**).

3. Find the **Enable computer use** toggle and turn it on.

4. Open Cowork or Claude Code in the desktop app and start a session.

5. Ask Claude to do something that involves an app on your computer. Claude will ask for permission to access the app before proceeding.

We’re sharing this early because it’s the kind of capability that gets better with real usage. If something doesn’t work as expected, use the in-app feedback button or reach out to [\[email protected\]](/cdn-cgi/l/email-protection#9ce9eff9eeeffdfaf9e8e5dcfdf2e8f4eef3ecf5ffb2fff3f1).

---
## configure-a-custom-opentelemetry-collector-for-office-agents-29202dd7c2.md

# Configure a custom OpenTelemetry collector for Office agents

August 5, 2026

Copy for LLM

You can route full audit telemetry from Office agents to your own OpenTelemetry (OTEL) collector. This gives your organization complete control over retention, encryption, and integration with your SIEM or observability platform.

This guide covers how to enable a custom collector, what data you'll receive, and the full span schema reference.

Custom OTEL collectors are available to Claude Enterprise organizations and to direct-provider deployments (Amazon Bedrock, Google Vertex AI, or a gateway).

## What you'll receive

When you configure a custom collector, Office agents send trace data covering every user turn. Each turn produces a tree of spans capturing the prompt, model calls, tool executions, file uploads, and context compaction events.

Your collector receives all span attributes, including those carrying user-generated content (prompt text, tool inputs and outputs, document URLs, and filenames). No attributes are redacted or filtered. Assistant response text is not included in the emitted span data. Your organization owns this data.

**Important:** Metrics aren't sent to custom collectors. The `office_agent.*` counter namespace routes to Anthropic only. However, every counter increment also appears as a span event on the active span, so the same signals are available in your traces.

Telemetry is sent via OTLP/HTTP to your endpoint at `{your_url}/v1/traces`. gRPC isn't supported because the add-in runs in an Office WebView.

------------------------------------------------------------------------

## Enable a custom collector

Setup differs depending on how your organization authenticates with Claude.

**Important:** When a custom endpoint is configured, telemetry goes exclusively to your collector. Spans aren't dual-sent to Anthropic.

### Claude Enterprise (OAuth) organizations

An organization administrator sets the collector endpoint in the Claude.ai admin console under **[Organization settings \> Office agents](https://claude.ai/admin-settings/office-agents)**. The setting applies organization-wide.

[TABLE]

The protocol must be HTTP-based OTLP. gRPC is rejected at configuration time.

### Direct provider deployments (Amazon Bedrock, Google Vertex AI, gateway)

For deployments that authenticate against your own model provider rather than Claude.ai, the collector endpoint is supplied through one of three configuration channels. All three use the same two keys.

**Recommended:** Use the **[claude-for-msft-365-install plugin](https://github.com/anthropics/financial-services/tree/main/claude-for-msft-365-install)** for Claude Code. It walks you through generating the manifest, registering Entra extension attributes, and standing up a bootstrap endpoint with `otlp_endpoint` and `otlp_headers` pre-wired. The three channels below are documented for reference and manual setup.

[TABLE]

If `otlp_endpoint` is unset or empty, no custom collector is configured and the add-in falls back to default behavior.

**Note:** These configuration channels apply to Microsoft Office deployments only. Google Workspace add-ins are configured separately.

**Channel 1: Manifest URL parameter**

Append the keys as query string parameters to the taskpane URL in your custom add-in manifest:

`https://<addin-host>/taskpane.html?otlp_endpoint=https://otel-collector.your-domain.com&otlp_headers=Authorization=Bearer%20<token>`

URL-encode the values. This applies the configuration to every user who installs the manifest.

**Channel 2: Azure Entra ID directory extension**

For per-user configuration, register the keys as Entra ID directory extension attributes and assign them via Microsoft Graph. The add-in reads them from the user's ID token using Nested App Authentication (NAA).

The claim names in the issued ID token follow Azure's directory extension format:

[TABLE]

Set these per-user with a Graph PATCH against the user object. Azure encodes directory extension values as single-element arrays in the ID token; the add-in unwraps them automatically. This channel requires `entra_sso=1` in the manifest URL parameters to enable NAA token acquisition.

**Channel 3: Bootstrap endpoint response**

If your deployment uses a bootstrap endpoint (a JSON endpoint your organization hosts that the add-in calls at startup), include the keys in the response body:

{
"otlp_endpoint": "https://otel-collector.your-domain.com",
"otlp_headers": "Authorization=Bearer <token>"
}

The bootstrap endpoint URL itself is configured via `bootstrap_url` in either the manifest URL parameters or an Entra `extn.bootstrap_url` claim. If an Entra ID token was acquired, it is passed to the bootstrap endpoint as a Bearer authorization header so your endpoint can authenticate the user before returning per-user configuration.

**Precedence**

When multiple channels supply a value, later channels override earlier ones: manifest parameters are read first, then Entra claims, then the bootstrap response. The bootstrap response wins.

If you haven't already, the fastest path is the **[claude-for-msft-365-install plugin](https://github.com/anthropics/financial-services/tree/main/claude-for-msft-365-install)**.

------------------------------------------------------------------------

## Deployment modes

Custom collector export is available on both deployment modes:

- Claude.ai Enterprise (OAuth): full audit trail including user identity (`user.email`, `user.account_uuid`, `organization.id`), MCP server metadata, and file upload records.

- Direct provider (Bedrock, Vertex AI, gateway): core audit trail (prompts, tool inputs and outputs, document URLs) but no user identity, no MCP metadata, and no file upload spans. User attribution requires correlating `session.id` against your own identity provider logs.

The core audit payload is identical in both modes. Direct provider deployments lack Claude.ai account context, so attributes derived from the Claude.ai user or organization profile are absent. See the `[claude.ai-only]` tags in the span schema below for the complete list.

## Surface and vendor labels

Every span includes two labels identifying which Office application and platform generated it. Use these as your primary dimensions for filtering and dashboards.

[TABLE]

------------------------------------------------------------------------

## Span reference

Each user turn produces a parent/child tree of up to five span types. Attributes marked \[content\] carry user-generated data; these form your audit payload. Attributes marked \[`claude.ai-only`\] are populated only when the user signs in with a Claude account; they are absent on Bedrock, Vertex AI, and gateway deployments. Absent attributes are omitted from the span entirely (no key with a null value).

The `file.upload` span and all `mcp.*` attributes are also \[`claude.ai-only`\], since file upload and MCP server connections are Claude platform features.

For direct provider deployments, user identity should be correlated via `session.id` and `document.url`, joined against your identity provider's session logs.

### Resource attributes

These attributes appear on every span:

[TABLE]

### agent.query (root span)

One span per user turn. This is the root of the span tree and carries session identity, document context, and MCP server status. SpanKind: INTERNAL.

[TABLE]

### agent.stream

One span per API call to Claude, as a child of the query span. SpanKind: CLIENT.

[TABLE]

**Note on prompt caching:** The add-in requests prompt caching unconditionally. The `cache_read_tokens` and `cache_creation_tokens` attributes are set from the provider's API response and are omitted when the response doesn't include them. Prompt caching is available for Claude Platform; as of this writing, Amazon Bedrock and Google Vertex AI don't yet return these fields through the client the add-in uses. When support lands on your provider, these attributes will begin appearing automatically.

### agent.tool_execution

One span per tool call, as a child of the stream span. SpanKind: INTERNAL. This is the primary record of what actions the model took against the document.

[TABLE]

**Note:** The `tool.accept_decision` attribute records how the permission decision was made: `manual` (the user approved this specific action), `auto_accept` (the user had previously granted standing approval), or `deferred` (the action was queued for later review). Use this to audit approval patterns across your organization.

### agent.compaction

One span per conversation auto-summarization, fired when context approaches the window limit. SpanKind: CLIENT.

[TABLE]

This span also carries `agent.surface`, `agent.vendor`, `session.id`, `user.email [claude.ai-only]`, `user.bucket [claude.ai-only]`, `office.platform`, and `office.version`, duplicated from the root span so you can query compaction events independently.

### file.upload \[claude.ai-only\]

One span per individual file upload, as a child of the query span. SpanKind: CLIENT. This span type only appears when users sign in with a Claude.ai account. File upload isn't available on direct provider deployments.

[TABLE]

------------------------------------------------------------------------

## Span events

Span events are timestamped markers attached to the spans above. They capture lifecycle transitions and counter-equivalent signals.

- `agent.query`: exception {exception.type}; file_upload {file_id, mime_type, content_category}

- `agent.stream`: first_token; stream_complete; stream_error {exception.type}

- `agent.tool_execution`: tool_init; tool_run; tool_result; tool_error {error_type}

- `agent.compaction`: compaction_start; compaction_complete; compaction_error {exception.type}

- `file.upload`: exception {exception.type}

Every internal product counter also records a span event with the same name on the currently active span, providing the equivalent of the metrics stream within your trace data. The `office_agent.token.usage` event is emitted on each `agent.stream` span, once per non-zero token type, with attributes {token_usage.type: input \| output \| cacheRead \| cacheCreation, token_usage.model, token_usage.token_count}. This mirrors the `*.token.usage` counter shape emitted by other Anthropic products, so a single collector can aggregate token cost across products by grouping on `service.name`.

------------------------------------------------------------------------

## Surface-specific behavior

The telemetry schema is consistent across all surfaces. These are the differences:

- Sheets (Excel): `agent.tool_execution` spans include `sheet.cells_read`, `sheet.cells_written`, and `sheet.cells_copied` attributes. `office_agent.cell_edit_collision_total` span events appear when a user is mid-cell-edit while a tool tries to write.

- Documents (Word): document-edit funnel events track the edit lifecycle: `office_agent.doc_edit_received_total`, `office_agent.doc_edit_parsed_total`, `office_agent.doc_edit_applied_total`, `office_agent.doc_proposed_edit_reviewed_total`. No `sheet.cells_*` attributes.

- Slides (PowerPoint): no surface-specific attributes or events beyond the common schema.

- Mail (Outlook): no surface-specific attributes or events beyond the common schema.

------------------------------------------------------------------------

## Reconstructing a user session

### Claude.ai deployments

- Filter spans by `user.email` (or `user.account_uuid`) and `session.id`.

- Order `agent.query` spans by start timestamp; each is one user turn.

- For each turn, `user.message` is the prompt and `document.url` is the file being worked on.

- Child `agent.tool_execution` spans, ordered by timestamp, are the actions taken: `tool.input` is what was attempted, `tool.output` is the result, `tool.accept_decision` records whether the user explicitly approved.

### Direct provider deployments

The add-in has no Claude.ai user identity in this mode, so spans carry no `user.email` or `user.account_uuid`. To attribute activity to a user:

- Filter spans by `session.id` to isolate one continuous add-in session.

- Use `document.url` to identify the file being worked on.

- Correlate the session against your identity provider's logs: Entra sign-in events, gateway access logs, or your bootstrap endpoint's request log (which receives the user's Entra ID token as a Bearer header).

- Once a session is attributed to a user, the per-turn reconstruction is identical: agent.query spans ordered by timestamp, with `user.message`, `tool.input`, `tool.output`, and `tool.accept_decision` providing the audit trail.

This produces a complete, ordered transcript of the interaction in both deployment modes.

---
## configuring-session-security-settings-7adbbc2e04.md

# Configuring session security settings

May 7, 2026

Copy for LLM

This feature is available to Admins and Owners of Enterprise plans and Console Admins.

Session duration controls allow Enterprise and Console Admins to set a maximum session length for all users in their organization. When enabled, users will need to sign in again after the specified period, even if they've been actively using Claude. This helps protect your organization by limiting how long a compromised session could remain valid.

## Enabling session length settings

### For Enterprise Admins

1. Log in to your Enterprise organization as an Admin or above.

2. Navigate to **[Organization settings \> Organization and access](https://claude.ai/admin-settings/organization)**.

3. Locate the **Session security** section.

4. Click “Enable” next to **Shortened session length**, then select a duration from the dropdown: 1 day, 7 days, 14 days, or 28 days.

5. Confirm your selection by clicking “Enable.”

### For Console Admins

1. Log in to your Console account as an Admin.

2. Navigate to **[Settings \> Organization and access](http://platform.claude.com/settings/organization)**.

3. Locate the **Session security** section.

4. Click “Enable” next to **Shortened session length**, then select a duration from the dropdown: 1 day, 3 days, or 7 days.

5. Confirm your selection by clicking “Enable.”

### What happens after enabling shortened session length?

- Existing sessions older than the selected duration will expire immediately.

- Other active sessions will expire no later than the selected duration.

- Users whose sessions expire will be directed to sign in again.

## Updating session duration

You can change the session duration at any time by selecting a new value from the dropdown. If you select a shorter duration:

- Sessions older than the new duration will expire immediately.

- Sessions scheduled to expire beyond the new duration will have their expiration shortened accordingly.

## Disabling session length settings

To disable session duration, select "Disable" next to **Shortened session length**. Existing active sessions will continue to expire at their scheduled time. New sessions will return to default behavior, where sessions remain active as long as the user stays active.

## Users in multiple organizations

If a user belongs to multiple organizations with different session duration settings, the shortest duration will be applied. For example, if a user is a member of Organization A (7-day limit) and Organization B (28-day limit), their sessions will expire after seven days. This is because a single session is used across all their organizations, so the most restrictive setting takes precedence.

---
## configuring-session-security-settings.md

4. Configuring session security settings

# Configuring session security settings

This feature is available to Admins and Owners of Enterprise plans and Console Admins.

## Enabling session length settings

### For Enterprise Admins

1. Log in to your Enterprise organization as an Admin or above.

2. Navigate to **[Organization settings \> Identity and access](https://claude.ai/admin-settings/identity)**.

3. Locate the **Session security** section.

4. Click “Enable” next to **Shortened session length**, then select a duration from the dropdown: 7 days, 14 days, or 28 days.

5. Confirm your selection by clicking “Enable.”

### For Console Admins

1. Log in to your Console account as an Admin.

2. Navigate to **[Settings \> Identity and access](http://platform.claude.com/settings/identity)**.

3. Locate the **Session security** section.

4. Click “Enable” next to **Shortened session length**, then select a duration from the dropdown: 1 day, 3 days, or 7 days.

5. Confirm your selection by clicking “Enable.”

### What happens after enabling shortened session length?

- Existing sessions older than the selected duration will expire immediately.

- Other active sessions will expire no later than the selected duration.

- Users whose sessions expire will be directed to sign in again.

## Updating session duration

You can change the session duration at any time by selecting a new value from the dropdown. If you select a shorter duration:

- Sessions older than the new duration will expire immediately.

- Sessions scheduled to expire beyond the new duration will have their expiration shortened accordingly.

## Disabling session length settings

## Users in multiple organizations

---
## consumer-terms-of-service-updates-4b571a2146.md

# Consumer Terms of Service Updates

March 16, 2026

Copy for LLM

This support article covers the changes to our [Consumer Terms of Service](https://www.anthropic.com/legal/consumer-terms) that take effect **May 1, 2024.** At Anthropic, we review our policies and terms to ensure we’re providing the highest level of protection and clarity for our users.

We encourage you to read our updated [Consumer Terms of Service](https://www.anthropic.com/legal/consumer-terms) in full, but at a high level, our latest updates provide clarity about how you can pay for and manage your subscription to our consumer services (like Claude Pro), if you purchased your subscription via an app.

In case you haven't heard, Claude is now available as an iOS app via the [Apple App Store](), putting the power of AI right in your pocket! And for businesses looking to supercharge their productivity, the new [Team plan](https://www.anthropic.com/claude) helps you collaborate with Claude like never before.

---
## consumer-terms-of-service-updates.md

# Consumer Terms of Service Updates

---
## cost-and-usage-reporting-in-the-claude-console-1b45b61cc1.md

# Cost and Usage Reporting in the Claude Console

March 16, 2026

Copy for LLM

**Note:** Usage and Cost reporting is visible to the following user roles: **Developer, Billing, and Admin**. See [Claude Console Roles and Permissions](https://support.claude.com/en/articles/10186004-claude-console-roles-and-permissions) for more information.

The Claude Console provides detailed cost and usage reporting to help you effectively manage your API usage and associated costs. This guide walks you through these features and how to use them.

## Accessing Cost and Usage Reports

Users with access to these reports can click into them on the left navigation menu on the Console:

------------------------------------------------------------------------

## Usage Reporting

The [Usage page](https://platform.claude.com/usage) offers a detailed breakdown of your API usage across different models and API keys.

### Key Features

- **Detailed Breakdown**: View usage data by model, date/time, and API key. Click into the bars on the bar chart for hour and minute granularity.

- **Flexible Filtering**: Use selectors to choose specific models, months, or API keys

- **Visual Representation**: A chart with input and output token counts.

- **Usage Statistics**: See total input and output tokens for your selected filters.

- **Rate-Limited Requests:** Review your requests that were blocked due to hitting rate limits.

- **Rate Limit Use:** Visualizations of input and output tokens per minute compared with the overall ITPM or OTPM rate limit.

- **CSV Export**: Download your usage data for further analysis or reporting.

### How to Use

1. Select the Workspace you want to view (or choose "All Workspaces").

2. Select the model you want to view (or choose "All Models").

3. Choose the month you're interested in (or narrow to a specific month/day).

4. Select an API key (or view data for all keys).

5. The chart and statistics will update based on your selections.

6. Use the export button to download a CSV of the displayed data.

### Rate Limit Use

The Usage page also includes a separate section displaying rate limit use per-model for input and output tokens. You can click the dropdown in the upper left corner of this section to change the model and view related rate limit metrics. These visualizations can be used to determine when you’re hitting peak use for your organization, which specific rate limits need to be increased, and how you can increase your caching rate.

**Rate Limit Use + Caching - Input Tokens:** This chart displays the hourly maximum number of uncached input tokens per minute (ITPM) alongside your cache rate (i.e. the percentage of input tokens read from the cache) and your current ITPM rate limit.

**Rate Limit Use - Output Tokens:** This chart displays the hourly maximum number of output tokens per minute (OTPM) alongside your current OTPM rate limit.

------------------------------------------------------------------------

## Cost Reporting

The [Cost page](https://platform.claude.com/cost) helps you understand your spending across different models.

### Key Features

- **Model-Specific Data**: View costs for individual models or all models combined.

- **Monthly Breakdown**: See costs for specific months.

- **Daily Cost Chart**: Visualize your spending over time.

- **Total Cost Statistics**: Get an overview of your total spending for the selected period, including web search and code execution costs.

- **CSV Export**: Download cost data for your records for further analysis.

### How to Use

1. Choose the Workspace you want to view costs for (or select "All Workspaces").

2. Choose the model you want to view costs for (or select "All Models").

3. Select the month you're interested in.

4. You can see the chart, token cost, and tool use costs, which will update based on your selections.

5. Use the export button to download a CSV of the cost data.

**Note**: Currently, it's not possible to break down usage or cost by individual users.

---
## crisis-helpline-support-in-claude-868e3770bf.md

# Crisis Helpline Support in Claude

March 16, 2026

Copy for LLM

*For medical emergencies, active injuries, or situations where you're in immediate physical danger, please contact emergency services (such as 911 in the US and Canada, or 112 in much of Europe) or go to your nearest emergency room.*

*This guidance was developed in partnership with ThroughLine.*

Claude may sometimes display crisis support resources in a banner or in its responses. These resources are designed to connect you with people who specialize in helping during difficult moments.

## Why am I seeing a crisis support message?

Claude may show a banner with connections to crisis support resources when your conversation touches on topics that may be related to emotional distress, suicidal thoughts, or self-harm. This banner is surfaced by a separate detection system, not from Claude directly.

These resources are there to make it easier to get help if you want it. You decide whether to use these resources, and there are no consequences for your account based on your decision to access resources.

If the banner doesn't feel relevant to your situation, that's completely okay - you can simply dismiss it. We'd rather offer support when it's not needed than miss an opportunity to help someone who could use it. After you see the banner, it won't appear again for about two hours, but you can always access crisis resources at [claude.findahelpline.com](http://claude.findahelpline.com) anytime.

## What is a crisis helpline?

Crisis helplines are free services staffed by trained personnel (who could be a volunteer, a counselor, or an individual who has been through similar situations) who provide support during tough times. You can usually reach them through multiple options (often phone, text, and/or online chat). Most operate around the clock, every day of the year, so help is available whenever you need it.

### What helplines offer

- **A caring ear:** Responders are there to hear what you're experiencing, without judgment or criticism.

- **Safety support:** If you're having thoughts of suicide or self-harm, they can work with you to create a plan for staying safe.

- **In-the-moment guidance:** They can help you work through overwhelming feelings and figure out next steps.

- **Connections to ongoing care:** They can point you toward community resources, support groups, or mental health services if you're looking for longer-term help.

- **Support for concerned loved ones:** If you’re worried about a friend or family member, helplines can help you think through how to be there for them.

You don't need to be at a breaking point to reach out. People contact helplines for all kinds of reasons - sometimes just to have someone listen.

## How to connect with a helpline

Contacting a helpline is straightforward, and there's no special preparation needed:

- **Choose your method:** Many helplines let you call, text, or chat online. You can check what's available and pick whichever feels most comfortable for you.

- **Speak freely:** Share what's on your mind however it comes out. There's no script to follow. Something like "I'm not sure where to start" works perfectly fine.

- **Feel free to let them guide the conversation:** The helpline responder will ask some questions to understand your situation and how they can best support you. You're always in control of what you share and where the conversation goes.

- **Share only what you're comfortable with:** You can remain anonymous. There’s no requirement to provide your name or any identifying details.

### A few things to keep in mind

- Crisis helplines provide immediate support, not ongoing therapy. However, they can help connect you with longer-term resources.

- For medical emergencies, active injuries, or situations where you're in immediate physical danger, please contact emergency services (911 in the US or Canada and 112 in much of Europe) or go to your nearest emergency room.

## Understanding who provides what

**Crisis helplines and Anthropic are entirely different organizations with different roles.** This distinction matters because it affects what kind of help you can get from each. It’s important to note that:

- **Helpline responders are not Anthropic employees.** They specialize in emotional and mental health support. Questions about Claude, your account, technical issues, or billing aren't something they can address.

- **Your Claude activity stays private.** Anthropic doesn't send your conversations to helplines. Helplines have no information about what you've discussed with Claude or any details about your account.

## Need help with Claude itself?

For anything related to Claude or your Anthropic account, here's where to go:

- **Support center:** Visit [support.anthropic.com](https://support.anthropic.com)

- **Account questions:** Check your account settings for help options

- **Feedback on responses:** The thumbs down button on any Claude response lets you flag concerns

## Frequently asked questions

We worked with ThroughLine to provide answers to common questions about using a crisis helpline.

### Is there a cost for using crisis helplines via ThroughLine?

Crisis helplines don't charge anything. Standard phone or texting rates from your carrier may apply, but the service itself is free.

### Where can I find a helpline?

If you're in the United States, the 988 Suicide & Crisis Lifeline is available by call or text. For other countries, a directory of services worldwide maintained by ThroughLine can be found at [claude.findahelpline.com](http://claude.findahelpline.com).

### Does talking to a helpline actually make a difference?

Studies consistently show that crisis counseling reduces distress and helps people feel more grounded. Responders are specifically trained to help during difficult moments, including supporting people through thoughts of self-harm.

### I'm not in a crisis. Is it still okay to call?

Absolutely. Many people reach out when they're stressed, lonely, or just having a hard day. You don't need to meet some threshold of "bad enough" to deserve support.

### Can I call about someone else I'm worried about?

Yes. Helplines support people who are concerned about others too. They can offer guidance on how to approach the situation and be helpful to the person you care about.

### Are there options besides phone calls?

Many helplines offer text and online chat alongside traditional phone support. You can use whatever channel feels most comfortable to you.

### Will the helpline contact my family or anyone else?

Generally, no. Conversations with support personnel are kept confidential. The main exception is if a responder believes there's an immediate, serious safety risk - and even then, they'll typically try to work with you first. Different helplines have slightly different policies, so feel free to ask when you connect.

### What about non-English speakers or people outside the United States?

Helplines exist in countries around the globe, many offering support in local languages. You can locate services in your region on [claude.findahelpline.com](http://claude.findahelpline.com).

### Is it okay to contact a helpline multiple times?

Yes, though some services have guidelines around frequency. Feel free to ask about their approach when you're connected.

### Does Anthropic send my Claude conversations to helplines?

No. Your interactions with Claude remain with Anthropic. Helpline personnel have no access to your messages, account information, or anything else related to your use of Claude. When you contact a helpline, that's a completely separate conversation.

### What if I have to wait to speak with someone?

High demand sometimes means wait times. If that happens, you can stay on the line, try reaching out via a different method (chat instead of phone, for instance), or contact an alternative helpline. [claude.findahelpline.com](http://claude.findahelpline.com) lists multiple options for most locations.

### I feel nervous about reaching out.

That's a common feeling. Helplines understand this and won't push you to share more than you're ready to. You control the pace and depth of the conversation. If it helps, you might start with something simple: "I'm feeling nervous about this call."

### How are the crisis resources in Claude selected?

We work with ThroughLine, an organization that curates and maintains a global directory of verified crisis helplines. The resources shown are selected based on your location to provide relevant, local support options.

---
## crisis-helpline-support-in-claude.md

# Crisis Helpline Support in Claude

*This guidance was developed in partnership with ThroughLine.*

Claude may sometimes display crisis support resources in a banner or in its responses. These resources are designed to connect you with people who specialize in helping during difficult moments.

## Why am I seeing a crisis support message?

## What is a crisis helpline?

### What helplines offer

- **A caring ear:** Responders are there to hear what you're experiencing, without judgment or criticism.

- **Safety support:** If you're having thoughts of suicide or self-harm, they can work with you to create a plan for staying safe.

- **In-the-moment guidance:** They can help you work through overwhelming feelings and figure out next steps.

- **Connections to ongoing care:** They can point you toward community resources, support groups, or mental health services if you're looking for longer-term help.

- **Support for concerned loved ones:** If you’re worried about a friend or family member, helplines can help you think through how to be there for them.

You don't need to be at a breaking point to reach out. People contact helplines for all kinds of reasons - sometimes just to have someone listen.

## How to connect with a helpline

Contacting a helpline is straightforward, and there's no special preparation needed:

- **Choose your method:** Many helplines let you call, text, or chat online. You can check what's available and pick whichever feels most comfortable for you.

- **Speak freely:** Share what's on your mind however it comes out. There's no script to follow. Something like "I'm not sure where to start" works perfectly fine.

- **Share only what you're comfortable with:** You can remain anonymous. There’s no requirement to provide your name or any identifying details.

### A few things to keep in mind

- Crisis helplines provide immediate support, not ongoing therapy. However, they can help connect you with longer-term resources.

## Understanding who provides what

## Need help with Claude itself?

For anything related to Claude or your Anthropic account, here's where to go:

- **Support center:** Visit [support.anthropic.com](https://support.anthropic.com)

- **Account questions:** Check your account settings for help options

- **Feedback on responses:** The thumbs down button on any Claude response lets you flag concerns

## Frequently asked questions

We worked with ThroughLine to provide answers to common questions about using a crisis helpline.

### Is there a cost for using crisis helplines via ThroughLine?

Crisis helplines don't charge anything. Standard phone or texting rates from your carrier may apply, but the service itself is free.

### Where can I find a helpline?

### Does talking to a helpline actually make a difference?

### I'm not in a crisis. Is it still okay to call?

Absolutely. Many people reach out when they're stressed, lonely, or just having a hard day. You don't need to meet some threshold of "bad enough" to deserve support.

### Can I call about someone else I'm worried about?

Yes. Helplines support people who are concerned about others too. They can offer guidance on how to approach the situation and be helpful to the person you care about.

### Are there options besides phone calls?

Many helplines offer text and online chat alongside traditional phone support. You can use whatever channel feels most comfortable to you.

### Will the helpline contact my family or anyone else?

### What about non-English speakers or people outside the United States?

Helplines exist in countries around the globe, many offering support in local languages. You can locate services in your region on [claude.findahelpline.com](http://claude.findahelpline.com).

### Is it okay to contact a helpline multiple times?

Yes, though some services have guidelines around frequency. Feel free to ask about their approach when you're connected.

### Does Anthropic send my Claude conversations to helplines?

### What if I have to wait to speak with someone?

### I feel nervous about reaching out.

### How are the crisis resources in Claude selected?

---
## data-usage-claude-code-docs-06b2864aaa.md

# Data usage

Learn about Anthropic’s data usage policies for Claude

> ## Documentation Index
>
> Fetch the complete documentation index at: <https://code.claude.com/docs/llms.txt>
>
> Use this file to discover all available pages before exploring further.

Data policies

Data training policy

**Consumer users (Free, Pro, and Max plans)**: We give you the choice to allow your data to be used to improve future Claude models. We will train new models using data from Free, Pro, and Max accounts when this setting is on (including when you use Claude Code from these accounts). **Commercial users**: (Team and Enterprise plans, API, 3rd-party platforms, and Claude Gov) maintain existing policies: Anthropic does not train generative models using code or prompts sent to Claude Code under commercial terms, unless the customer has chosen to provide their data to us for model improvement (for example, the [Developer Partner Program](https://support.claude.com/en/articles/11174108-about-the-development-partner-program)).

Development Partner Program

If you explicitly opt in to methods to provide us with materials to train on, such as via the [Development Partner Program](https://support.claude.com/en/articles/11174108-about-the-development-partner-program), we may use those materials provided to train our models. An organization admin can expressly opt-in to the Development Partner Program for their organization. Note that this program is available only for Anthropic first-party API, and not for Bedrock or Vertex users.

Feedback using the `/feedback` command

If you choose to send us feedback about Claude Code using the `/feedback` command, we may use your feedback to improve our products and services. Transcripts shared via `/feedback` are retained for 5 years.

Session quality surveys

When you see the “How is Claude doing this session?” prompt in Claude Code, responding to this survey, including selecting “Dismiss”, records only your rating. We do not collect or store any conversation transcripts, inputs, outputs, or other session data as part of the rating prompt itself. Unlike thumbs up/down feedback or `/feedback` reports, this session quality survey is a simple product satisfaction metric. After the rating prompt, you may see a separate follow-up asking “Can Anthropic look at your session transcript to help us improve Claude Code?”. This is an optional second step distinct from the rating:

- **Yes**: uploads your conversation transcript, any subagent transcripts, and the raw session log file from disk to Anthropic. Known API key and token patterns are redacted before upload. Source code, file contents, and other conversation content are uploaded as-is. Shared transcripts are retained for up to 6 months.
- **No**: declines without sending anything
- **Don’t ask again**: declines and stops this follow-up from appearing in future sessions

Nothing is uploaded unless you explicitly select **Yes**. Organizations with [zero data retention](/docs/en/zero-data-retention), or where product feedback is disabled by organization policy, or where `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` is set, never see this follow-up. Your responses to this survey, including session transcripts submitted after the rating prompt, do not impact your data training preferences and cannot be used to train our AI models. To disable these surveys, set `CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1`. The survey is also disabled when `DISABLE_TELEMETRY`, `DO_NOT_TRACK`, or `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` is set. Organizations that block nonessential traffic but capture survey responses through their own [OpenTelemetry collector](/docs/en/monitoring-usage) can opt the survey back in by setting `CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL=1`. The survey then logs ratings to the configured collector only. The transcript-share follow-up and all other Anthropic-bound feedback traffic stay disabled. To control frequency instead of disabling, set [`feedbackSurveyRate`](/docs/en/settings#available-settings) in your settings file to a probability between `0` and `1`.

Data retention

Anthropic retains Claude Code data based on your account type and preferences. **Consumer users (Free, Pro, and Max plans)**:

- Users who allow data use for model improvement: 5-year retention period to support model development and safety improvements
- Users who don’t allow data use for model improvement: 30-day retention period
- Privacy settings can be changed at any time at [claude.ai/settings/data-privacy-controls](https://claude.ai/settings/data-privacy-controls).

**Commercial users (Team, Enterprise, and API)**:

- Standard: 30-day retention period
- [Zero data retention](/docs/en/zero-data-retention): available for Claude Code on Claude for Enterprise. ZDR is enabled on a per-organization basis; each new organization must have ZDR enabled separately by your account team
- Local caching: Claude Code clients store session transcripts locally in plaintext under `~/.claude/projects/` for 30 days by default to enable session resumption. Adjust the period with `cleanupPeriodDays`. See [application data](/docs/en/claude-directory#application-data) for what’s stored and how to clear it.

You can delete individual Claude Code on the web sessions at any time. Deleting a session permanently removes the session’s event data. For instructions on how to delete sessions, see [Delete sessions](/docs/en/claude-code-on-the-web#delete-sessions). Learn more about data retention practices in our [Privacy Center](https://privacy.anthropic.com/). For full details, please review our [Commercial Terms of Service](https://www.anthropic.com/legal/commercial-terms) (for Team, Enterprise, and API users) or [Consumer Terms](https://www.anthropic.com/legal/consumer-terms) (for Free, Pro, and Max users) and [Privacy Policy](https://www.anthropic.com/legal/privacy).

Data access

For all first party users, you can learn more about what data is logged for [local Claude Code](#local-claude-code-data-flow-and-dependencies) and [remote Claude Code](#cloud-execution-data-flow-and-dependencies). [Remote Control](/docs/en/remote-control) sessions follow the local data flow since all execution happens on your machine. Note for remote Claude Code, Claude accesses the repository where you initiate your Claude Code session. Claude does not access repositories that you have connected but have not started a session in.

Local Claude Code: Data flow and dependencies

The diagram below shows how Claude Code connects to external services during installation and normal operation. Solid lines indicate required connections, while dashed lines represent optional or user-initiated data flows. Claude Code runs locally. To interact with the LLM, Claude Code sends data over the network. This data includes all user prompts and model outputs, encrypted in transit via TLS 1.2+. Claude Code is compatible with most popular VPNs and LLM proxies. Encryption at rest depends on your model provider:

| Provider | Encryption at rest |
|------------------------|--------------------------------------------------------------------------------------------------------------------------------------------|
| Anthropic API | Infrastructure-level disk encryption (AES-256). Enable [Zero Data Retention](/docs/en/zero-data-retention) for no server-side persistence. |
| Amazon Bedrock | AES-256 with AWS-managed keys. Customer-managed keys available via AWS KMS. |
| Google Cloud Vertex AI | Google-managed encryption keys. CMEK available. |
| Microsoft Foundry | Requests route to Anthropic infrastructure with AES-256 disk encryption. |

Claude Code is built on Anthropic’s APIs. For details on API security controls, including API logging procedures, see the compliance artifacts in the [Anthropic Trust Center](https://trust.anthropic.com).

Cloud execution: Data flow and dependencies

When using [Claude Code on the web](/docs/en/claude-code-on-the-web), sessions run in Anthropic-managed virtual machines instead of locally. In cloud environments:

- **Code and data storage:** Your repository is cloned to an isolated VM. Code and session data are subject to the retention and usage policies for your account type (see Data retention section above)
- **Credentials:** GitHub authentication is handled through a secure proxy; your GitHub credentials never enter the sandbox
- **Network traffic:** All outbound traffic goes through a security proxy for audit logging and abuse prevention
- **Session data:** Prompts, code changes, and outputs follow the same data policies as local Claude Code usage

For security details about cloud execution, see [Security](/docs/en/security#cloud-execution-security).

Telemetry services

Claude Code connects from users’ machines to Anthropic to log operational metrics such as latency, reliability, and usage patterns. This logging does not include any code or file paths. Data is encrypted in transit and at rest. To opt out of telemetry, set the `DISABLE_TELEMETRY` environment variable. Claude Code connects from users’ machines to Sentry for operational error logging. The data is encrypted in transit using TLS and at rest using 256-bit AES encryption. Read more in the [Sentry security documentation](https://sentry.io/security/). To opt out of error logging, set the `DISABLE_ERROR_REPORTING` environment variable. When you run the `/feedback` command, a copy of your conversation history including code is sent to Anthropic. Before submitting, you choose how much history to include: the current session only, which is the default, or also other sessions from the same project over the last 24 hours or 7 days. The data is encrypted in transit via TLS. Optionally, a GitHub issue is created in the public repository. To opt out, set the `DISABLE_FEEDBACK_COMMAND` environment variable to `1`. When you use a third-party provider such as Bedrock or Vertex, or have no Anthropic credentials configured, `/feedback` writes the report to a local archive under `~/.claude/feedback-bundles/` instead of sending it to Anthropic. Known API key and token patterns are redacted before the archive is written. Nothing leaves your machine until you send that file to your Anthropic account representative or attach it to a support request.

Default behaviors by API provider

By default, error reporting, telemetry, and bug reporting are disabled when using Bedrock, Vertex, Foundry, or Claude Platform on AWS. Session quality surveys and the WebFetch domain safety check are exceptions and run regardless of provider. You can opt out of all non-essential traffic, including surveys, at once by setting `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC`. This variable does not affect the WebFetch check, which has its own opt-out. Here are the full default behaviors:

[TABLE]

All environment variables can be checked into `settings.json` (see [settings reference](/docs/en/settings)). As of v2.1.126, when a host platform sets `CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST`, metrics default to on for Vertex, Bedrock, and Foundry, and follow the standard `DISABLE_TELEMETRY` opt-out. Sentry error reporting and `/feedback` reports remain off by default on those providers.

WebFetch domain safety check

Before fetching a URL, the WebFetch tool sends the requested hostname to `api.anthropic.com` to check it against a safety blocklist maintained by Anthropic. Only the hostname is sent, not the full URL, path, or page contents. Results are cached per hostname for five minutes. This check runs regardless of which model provider you use and is not affected by `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC`. If your network blocks `api.anthropic.com`, WebFetch requests fail until you either allowlist the domain or set `skipWebFetchPreflight: true` in [settings](/docs/en/settings). Disabling the check means WebFetch attempts to retrieve any URL without consulting the blocklist, so combine it with [`WebFetch` permission rules](/docs/en/permissions#webfetch) if you need to restrict which domains Claude can reach.

---
## exceptions-to-our-usage-policy-ff5ed2bd47.md

# Exceptions to our Usage Policy

March 16, 2026

Copy for LLM

Anthropic maintains a comprehensive Usage Policy to ensure the responsible deployment of our AI systems. Anthropic may enter into contracts with government customers that tailor use restrictions to that customer’s public mission and legal authorities if, in Anthropic’s judgment, the contractual use restrictions and applicable safeguards are adequate to mitigate the potential harms addressed by this Usage Policy.

For example, with carefully selected government entities, we may allow foreign intelligence analysis in accordance with applicable law. All other use restrictions in our Usage Policy, including those prohibiting use for disinformation campaigns, the design or use of weapons, censorship, domestic surveillance, and malicious cyber operations, remain.

At this time, this policy only applies to models that are at AI Safety Level 2 (ASL-2) under our Responsible Scaling Policy (RSP).

**What government entities are eligible for Usage Policy modifications?**

Our evaluation of whether to tailor use restrictions to the mission and legal authorities of a government entity aims to balance enabling beneficial uses of our products and services with mitigating potential harms, and includes:

- Our assessment of the models’ suitability for the proposed use cases.

- The legal authorities of the agency in question.

- The extent of the agency's willingness to engage in ongoing dialogue with Anthropic.

- The safeguards in place to prevent misuse and mitigate risks of mistakes.

- The degree of independent and democratic oversight of the organizations and their uses of AI technologies, including legislative or regulatory constraints and other relevant public commitments.

---
## exceptions-to-our-usage-policy.md

# Exceptions to our Usage Policy

At this time, this policy only applies to models that are at AI Safety Level 2 (ASL-2) under our Responsible Scaling Policy (RSP).

**What government entities are eligible for Usage Policy modifications?**

- Our assessment of the models’ suitability for the proposed use cases.

- The legal authorities of the agency in question.

- The extent of the agency's willingness to engage in ongoing dialogue with Anthropic.

- The safeguards in place to prevent misuse and mitigate risks of mistakes.

- The degree of independent and democratic oversight of the organizations and their uses of AI technologies, including legislative or regulatory constraints and other relevant public commitments.

---
## getting-started-with-local-agent-mode.md

# Get started with Cowork

This article explains how to use **[Cowork](https://claude.com/product/cowork)**, a research preview that brings Claude Code's agentic capabilities to Claude Desktop for knowledge work beyond coding.

## Availability

Cowork is available as a research preview for paid plans (Pro, Max, Team, Enterprise) on:

- **Claude Desktop for macOS**

- **[Click here](https://claude.ai/api/desktop/darwin/universal/dmg/latest/redirect)** to download

- **Claude Desktop for Windows** (x64 only)

- Cowork requires the latest version of Claude for Windows. Download or update at **[claude.com/download](https://claude.com/download)**.

- Windows arm64 is not supported.

------------------------------------------------------------------------

## What is Cowork?

Cowork uses the same agentic architecture that powers Claude Code, now accessible within Claude Desktop and without opening the terminal. Instead of responding to prompts one at a time, Claude can take on complex, multi-step tasks and execute them on your behalf.

With Cowork, you can describe an outcome, step away, and come back to finished work—formatted documents, organized files, synthesized research, and more. With scheduled tasks, Claude can complete work for you automatically, which isn't possible in regular chats outside of Cowork. With the introduction of projects in Cowork, you can organize related tasks into persistent, self-contained workspaces with their own files, links, instructions, and memory, making Cowork more powerful for recurring or long-running work.

**Important:**

- Cowork is a research preview with unique risks due to its agentic nature and internet access.

- Cowork respects your current network egress permissions.

- Network egress permissions don't apply to the web search tool.

- Cowork stores conversation history locally on your computer, so is not subject to Anthropic's **[data retention timeframe](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data)**.

- Cowork activity is not captured in Audit Logs, Compliance API, or Data Exports. Do not use Cowork for regulated workloads.

- Please review **[Use Cowork safely](https://support.claude.com/en/articles/13364135-using-cowork-safely)** for more information.

For important limitations and considerations for Team and Enterprise organizations using Cowork, see **[Cowork for Team and Enterprise plans](../17-Billing-Plans/cowork-for-team-and-enterprise-plans.md)**.

### Key capabilities

- **Direct local file access:** Claude can read from and write to your local files without manual uploads or downloads.

- **Sub-agent coordination:** Claude breaks complex work into smaller tasks and coordinates parallel workstreams to complete them.

- **Professional outputs:** Generate polished deliverables like Excel spreadsheets with working formulas, PowerPoint presentations, and formatted documents.

- **Long-running tasks:** Work on complex tasks for extended periods without conversation timeouts or context limits interrupting your progress.

- **Scheduled tasks:** Create and save tasks that you can have Claude run on-demand, or automatically on a cadence of your choosing.

- **Spreadsheets and presentations:** Cowork can produce spreadsheets and slides that can be further edited with Claude for Excel and Powerpoint.

- **Projects:** Group related tasks into separate workspaces with their own files, context, instructions, and memory. See **[Organize your tasks with projects in Cowork](https://support.claude.com/en/articles/14116274-organize-your-tasks-with-projects-in-cowork).**

- **Mobile access for Pro and Max:** Message Claude from your phone and get results delivered back to the same conversation. Claude works on your desktop using your local files and connectors — you just don't have to be sitting in front of it. See **[Assign tasks to Claude from anywhere in Cowork](../21-Account-Support/assign-tasks-to-claude-from-anywhere-in-cowork.md)**.

------------------------------------------------------------------------

## How Cowork runs your tasks

Cowork runs directly on your computer, giving Claude access to the files you choose to share. Code runs safely in an isolated space, but Claude can make real changes to your files.

When you start a task in Cowork, Claude:

1. Analyzes your request and creates a plan.

2. Breaks complex work into subtasks when needed.

3. Executes work in a virtual machine (VM) environment.

4. Coordinates multiple workstreams in parallel if appropriate.

5. Delivers finished outputs directly to your file system.

You maintain visibility into what Claude is planning and doing throughout the process so you can steer when it matters, or let Claude run independently.

------------------------------------------------------------------------

## Get started

### Requirements

- **Claude Desktop app:** Cowork requires the **[desktop app](https://support.claude.com/en/articles/10065433-installing-claude-desktop)** for macOS or Windows and is not available on web or mobile.

- **Paid Claude subscription:** This research preview is available to paid Claude plans (Pro, Max, Team, Enterprise) only.

- **Active internet connection:** Required throughout the session.

### Access Cowork

1. Open Claude Desktop.

2. Look for the mode selector that includes "Chat" and the Cowork tab.

3. Click the "Cowork" tab to switch modes to "Tasks".

4. Describe the task you want Claude to complete.

5. Review Claude's approach, then let it run.

**Note:** The Claude Desktop app must remain open while Claude is working. If you close the app, your session will end.

## What to expect during a task

When Claude is working on a task in Cowork:

- **Progress indicators** show what Claude is doing at each step.

- **Transparency:** Claude surfaces its reasoning and approach so you can follow along.

- **Steering:** You can jump in to course-correct or provide additional direction mid-task.

- **Parallel work:** For complex tasks, Claude may coordinate multiple sub-agents working simultaneously.

- **Deletion protection:** When using Cowork, Claude requires your explicit permission before permanently deleting any files. You will see a permission prompt and will need to select "Allow" before Claude is allowed to perform deletion tasks.

Tasks can run for extended periods depending on complexity. You can monitor progress or step away and return when Claude finishes.

------------------------------------------------------------------------

## Add global and folder instructions

### Global instructions

You can give Claude standing instructions that apply to every Cowork session. Use this to specify your preferred tone, output format, or background on your role.

To set global instructions:

1. Navigate to Settings \> Cowork within Claude Desktop.

2. Click "Edit" next to **Global instructions**.

3. Type your instructions in the text box and click "Save":

### Folder instructions

Folder instructions add project-specific context to Cowork when you select a local folder. Claude can also update these on its own during a session.

------------------------------------------------------------------------

## Cowork plugins

Plugins customize how Claude works for your role, team, and company in Cowork. Each one bundles skills, connectors, and sub-agents into a single package. For details on finding, installing, and customizing plugins, see **[Use plugins in Cowork](../15-Claude-AI-Features/use-plugins-in-cowork.md)**.

------------------------------------------------------------------------

## Schedule recurring tasks

You can set up tasks that Claude runs automatically or on demand. To schedule a task, type `/schedule` in any Cowork task. You can also click "Scheduled" in the left sidebar to view, create, and manage your scheduled tasks.

Scheduled tasks only run while your computer is awake and the Claude Desktop app is open.

For more in-depth details, see **[Schedule recurring tasks in Cowork](../15-Claude-AI-Features/schedule-recurring-tasks-in-cowork.md)**.

------------------------------------------------------------------------

## Usage limits

Working on tasks with Cowork consumes more of your usage allocation than chatting with Claude. This is because complex, multi-step tasks are compute-intensive and require more tokens to execute.

If you find yourself hitting usage limits frequently when using the Cowork research preview, consider:

- Batching related work into single sessions.

- Using standard chat for simpler tasks that don't require file access or extended execution.

- Monitoring your individual usage in **[Settings \> Usage](http://claude.ai/settings/usage)**.

See **[Usage limit best practices](https://support.claude.com/en/articles/9797557-usage-limit-best-practices)** for more information.

------------------------------------------------------------------------

## Example use cases

Cowork is designed for complex, multi-step work that benefits from file access and extended execution time. Here are some examples:

### File and document management

- **Organize files:** "Organize my Downloads folder by type and date" — Claude can sort hundreds of files into categorized folders.

- **Process receipts:** Drop receipts in a folder and ask Claude to create a formatted expense report.

- **Batch rename:** Rename files with consistent patterns like YYYY-MM-DD formatting.

### Research and analysis

- **Research synthesis:** Combine information from web searches, articles, papers, and notes into coherent reports or summaries.

- **Transcript analysis:** Extract themes, key points, and action items from meeting notes, interviews, or lecture recordings.

- **Personal knowledge synthesis:** Analyze your notes, journals, or research files to surface patterns, themes, and connections you might have missed.

### Document creation

- **Spreadsheets with formulas:** Generate Excel files with working VLOOKUP, conditional formatting, and multiple tabs—not just CSVs that need fixing.

- **Presentations:** Create slide decks from rough notes or meeting transcripts.

- **Reports from messy inputs:** Turn voice memos and scattered notes into polished documents.

### Data and analysis

- **Statistical analysis:** Outlier detection, cross-tabulation, and time-series analysis on your data files.

- **Data visualization:** Generate charts using your data.

- **Data transformation:** Clean, transform, and process datasets.

------------------------------------------------------------------------

## Permissions and security

Cowork runs in a virtual machine (VM) on your computer. This provides several security benefits:

- **Controlled environment:** Claude operates within defined boundaries, with controlled file and network access.

- **Isolation:** The VM environment is separate from your main operating system.

**Important:** While the VM provides isolation, Claude does have access to local files you grant it permission to access. Review Claude's planned actions before allowing it to proceed, especially when working with sensitive files.

### Permissions

Permissions work the same as for chat. You control:

1. Which **[MCPs you connect to Claude](https://claude.ai/settings/connectors)** and how often they ask for permission.

2. **[Claude's internet access](https://claude.ai/settings/capabilities)**

Please carefully assess how much you trust an MCP or website before extending access beyond Claude's default settings.

------------------------------------------------------------------------

## Current limitations

Cowork is a feature preview, which means some capabilities are not yet available:

- **Memory in projects only:** Memory is supported within projects but is not retained across standalone Cowork sessions.

- **No chat or artifact sharing:** Sessions cannot be shared with others.

- **Desktop app required:** Cowork runs on your desktop computer via the Claude Desktop app. Pro and Max users can also message Claude from the mobile app while your desktop stays active. See Assign tasks to Claude from anywhere in Cowork for details.

- **Session persistence:** The Claude Desktop app must remain open and your computer must be awake for Claude to work on tasks. If you close the app or your computer goes to sleep, active tasks will stop.

We're iterating on Cowork based on feedback. If you encounter issues or have suggestions, use the feedback button in the app to share feedback with our team.

------------------------------------------------------------------------

## Troubleshooting

### I'm seeing "Setting up Claude's workspace" when I start Cowork; what does this mean?

This message is expected and indicates that Cowork is updating to the most recent version to apply any fixes and improvements.

### Claude stopped working on my task

Ensure the Claude Desktop app was open throughout the entire task. If the app was closed or your computer went to sleep, the session may have ended.

### I'm hitting usage limits quickly

Cowork consumes more usage than standard chat. Try using standard chat for simpler tasks and reserve Cowork for complex, multi-step work that benefits from file access.

### Files aren't appearing where expected

Check that you've granted Claude the appropriate file access permissions. Review the output location Claude specified when completing the task.

---
## holiday-2025-usage-promotion-843f18cc62.md

# Holiday 2025 Usage Promotion

March 16, 2026

Copy for LLM

We’re giving Pro and Max subscribers a holiday gift—2x usage limits from December 25 through December 31. During this period, your plan’s usage limits will be doubled, giving you more room to explore Claude’s capabilities as you plan for the new year.

## Who is eligible?

To be eligible for this promotion, you must meet the following criteria:

- You have an active Pro, Max 5x, or Max 20x subscription.

- You are an individual subscriber (this promotion does not apply to Team or Enterprise plans).

- Both existing subscribers and new users who sign up for Pro or Max during the promotional period are eligible.

## What’s included in this promotion?

During the promotional period, you’ll receive:

- A 2x increase to your five-hour usage limits.

- A 2x increase to your weekly usage caps.

All weekly caps will be reset at the start of the promotional period (December 25 at 12:00 AM UTC).

This promotion applies to usage on [claude.ai](http://claude.ai), Claude Code, and Claude in Chrome.

## When does this promotion run?

This is a one-week holiday promotion running from Wednesday, December 25, 2025 at 12:00 AM UTC through Wednesday, December 31 at 11:59 PM UTC.

## What happens after the promotional period ends?

After the promotional period ends, your usage limits will return to their standard levels based on your subscription plan. There is no additional charge—you’ll simply continue using Claude as part of your existing subscription.

## Terms and Conditions

Additional usage is available from December 25, 2025 at 12:00 AM UTC through December 31, 2025 at 11:59 PM UTC. The increased usage limits apply to the Claude web app, Claude Code, and Claude in Chrome only. This promotion cannot be combined with other offers or promotions. Available to individual Pro, Max 5x, and Max 20x subscribers only. Team and Enterprise plans are not eligible.

---
## how-can-i-create-and-manage-projects-e54efbf170.md

# How can I create and manage projects?

Copy for LLM

Projects are available to all users, including those with free Claude accounts. Free users can create a maximum of five projects.

## How to create a project

1. Hover over the left side of your account and click “Projects,” or navigate directly to this section: **[claude.ai/projects](https://claude.ai/projects)**.

2. Click "+ New Project" in the upper right corner.

3. Give your project a name and description (note that Claude will not have access to these details).

4. If you are on a Team or Enterprise plan, choose the visibility settings:

- **Keep it private:** Only you and members you invite can view or use this project.

- **Share with your broader organization:** Everyone in your organization can view and use this project (where enabled by administrators).

Once your project is set up, you can initiate chats with Claude within the project.

## Add content to project knowledge

You'll find the project knowledge base on the right side of your project's main page. Anything you upload to this space will be used across all of your chats within that project.

### Add content to your project knowledge base

1. Click on the "+" button to add content to the project.

2. Upload relevant documents, text files, or code snippets.

3. Claude will process this information and use it as context in your chats within the project.

4. If you are using a paid Claude plan, when your project knowledge approaches the context window limit, **[Claude will automatically enable RAG mode to expand your project's capacity](https://support.anthropic.com/en/articles/11473015-retrieval-augmented-generation-rag-for-projects)**.

### Add project instructions to your project knowledge base

1. Click on "Set project instructions."

2. Add instructions for how you'd like Claude to behave and respond and click "Save instructions."

3. Claude will use these instructions for all the chats within the project.

**Note:** Context is not shared across chats within a project unless the information is added into the project knowledge base.

------------------------------------------------------------------------

## Share projects

If you're on a Team or Enterprise plan, you can share projects with other users in your organization, unless your admin has turned off project sharing. See **[If you can't share projects](https://support.claude.com/en/articles/9519189-manage-project-visibility-and-sharing#h_d5a68ef397)** for what to expect.

### To share a project

1. Open the project you want to share.

2. Click the "Share project" button to the right of the project name.

3. Add individual members using their name or email address, or copy and paste a list of email addresses for bulk sharing.

4. Select the appropriate permission level for each member:

- **Can view:** Members can see project contents, knowledge, and instructions, and chat within the project, but cannot edit it.

- **Can edit:** Members can modify project instructions and knowledge, update member settings, and actively contribute to the project.

5. Click "Share" to send invitations.

### Manage shared project access

- **To change permissions:** Open the sharing menu and select a new role for the member.

- **To view who has access:** The sharing menu shows all current members and their permission levels.

- **To remove access:** The project creator can click on the member's current role and select "Remove access."

### Find shared projects

Your **[Projects page](https://claude.ai/projects)** has three tabs: "Your projects," "Organization," and "Shared with you." "Your projects" lists the projects you created, and "Organization" lists projects that other members have shared with your organization. If you can't find a project, check all three tabs.

------------------------------------------------------------------------

## Star projects for quick access

Starring a project allows for quick access from your projects and chats list, visible when hovering over the left side of your account. You can star a project from two different areas:

### From the “Projects” page

1. Locate the project on the “Projects” page.

2. Click on the three dots ("..." button) on the upper right corner of the project.

3. Select "Star" from the menu that appears.

### From the project

1. Open the relevant project.

2. Click on the star icon in the upper right corner.

3. The project will now appear in your starred items in the left side panel of your account.

------------------------------------------------------------------------

## How to move chats into projects

You can move a standalone chat into a project by clicking on the dropdown arrow next to the chat name, then “Add to project”:

Browse or search for the correct project in the **Move chat** modal that appears, then click on it to move the chat.

You can also remove chats from projects, or move them between projects, using the same dropdown menu within the chat:

------------------------------------------------------------------------

## Move chats to manage Claude’s memory

Memory is on by default for Free, Pro, and Max plans on the web, Claude Desktop, and Claude Mobile. On Team and Enterprise plans, memory is available when an owner has enabled it.

For Team and Enterprise plans using Claude’s memory, the ability to move chats into and out of projects allows you to manage what’s included in Claude’s memory. Each project has its own memory, kept separate from your non-project chats. For example, if you accidentally start an unrelated chat in a project and need to remove it from the project-specific memory summary, you can click “Remove from project” so it will be included in Claude’s non-project memory instead.

Refer to our article on chat search and memory for more information: **[What is Claude’s memory?](https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context#h_82126ebcc9)**

------------------------------------------------------------------------

## How to archive projects

Archiving is a way to organize your projects and declutter your project list. It's particularly useful for completed projects or those you're not actively working on but may need to reference in the future.

**Important:** Archiving a project doesn't reset its sharing permissions or remove members. All members, permission levels, and project knowledge are preserved, and everything is restored exactly as it was when you unarchive the project. To revoke someone's access, remove them in the project's sharing settings.

Here's what you need to know about archiving:

- Archived projects will appear at the bottom of your main projects list.

- You can still access conversations from archived projects.

- There is no way to delete archived projects at this time (you must unarchive it first).

### To archive a project, follow these steps

1. Navigate to the project you want to archive.

2. Click on the three dots ("..." button) in the upper right corner.

3. Confirm that you want to archive the project.

### How to access archived projects

You can access an archived project in the archived projects tab on the **[Projects page](https://claude.ai/projects)**.

## How to unarchive a project

There are two ways to make archived project active again:

### From the archived projects tab

1. Locate the archived project.

2. Click on the three dots ("..." button) in the upper right corner.

3. Select "Unarchive" from the menu that appears.

### From the project

1. Navigate to the archived project.

2. Click on the three dots ("..." button) in the upper right corner.

3. Confirm that you want to unarchive the project.

------------------------------------------------------------------------

## Delete projects

**Note:** It's not possible to delete archived projects. To see the option to delete an archived project, you must unarchive it first.

### From the “Projects” page

1. Locate the project on the **[Projects page](https://claude.ai/projects)**.

2. Click on the three dots ("..." button) on the upper right corner of the project.

3. Select "Delete" from the menu that appears.

4. Confirm deletion in the pop-up by clicking "Yes, delete."

### From the project

1. Navigate to the project

2. Click on the three dots ("..." button) next to the star.

3. Select "Delete."

4. Confirm deletion in the pop-up by clicking "Yes, delete."

---
## how-do-i-pay-for-my-api-usage-96667569cd.md

# How do I pay for my Claude API usage?

August 19, 2026

Copy for LLM

This article explains how billing works for the Claude API, the playground, and Claude Code when you use them through a Claude Console account. Most organizations pay with prepaid usage credits. Organizations with an invoicing arrangement are billed monthly instead.

## Prepaid usage credits

Claude API and playground usage is billed through prepaid usage credits. Buy credits before you use the API, and they're applied to your usage according to our current **[pricing](https://claude.com/pricing#api)**. Credits cover API access, playground usage, and Claude Code.

You're billed only for successful API calls and completed tasks. Failed requests aren't charged.

**Note:** If your client disconnects or times out in the middle of a request that was on track to succeed, that request is still charged.

If you run out of credits, you can no longer call the API or use the playground until you add more.

## Buy credits

1. Log in to the Console with an Admin or Billing role.

2. Navigate to **[Settings \> Billing](https://platform.claude.com/settings/billing)**.

3. Click "Buy credits."

4. Enter the amount of credits you want to purchase and confirm.

Purchased credits are available immediately. Your organization's available credit balance and credit usage are shown on the same **Billing** page.

## Set up auto-reload

Auto-reload purchases additional credits automatically when your balance falls below a threshold you set.

1. On the **Billing** page, click "Edit" in the **Auto-reload** section.

2. Toggle auto-reload on or off.

3. If auto-reload is on, set the minimum balance that triggers a purchase and the amount to reload to.

## Credit expiration and refunds

Purchased credits are subject to our **[Credit Terms](https://www.anthropic.com/legal/credit-terms)**. Credits expire one year from the purchase date, and the expiration date can't be extended. Expired credits appear in your **Invoice history** on the Billing page. All credit purchases are non-refundable.

Learn more about how credit usage relates to rate limits in the **[Claude API docs](https://platform.claude.com/docs/en/api/rate-limits)**.

## Monthly invoicing

Some organizations are billed monthly in arrears instead of buying credits upfront. This applies to Console organizations with an invoicing arrangement set up through our Sales team.

On monthly invoicing, we aggregate your usage across API calls, Console usage, and any other services associated with your account, and bill it at our standard **[pay-as-you-go pricing](https://claude.com/pricing#api)**. At the end of each calendar month, you'll receive an invoice from Stripe. Enter your payment details in Stripe to pay it.

You can view charges for the current billing period on the **[Billing](https://platform.claude.com/settings/billing)** page in your Console settings.

If you need custom rate limits, monthly invoicing, or hands-on support, **[contact our Sales team](https://claude.com/contact-sales)**.

## Update your Console payment method

1. Log in to the Console with an Admin or Billing role.

2. Navigate to **[Settings \> Billing](https://platform.claude.com/settings/billing)**.

3. Click the pencil icon next to your current payment method.

4. Enter your new card details in the **Update payment method** modal, then click "Update."

---
## how-do-i-pay-for-my-api-usage.md

4. How do I pay for my Claude API usage?

# How do I pay for my Claude API usage?

Claude API and Workbench usage is billed via prepaid "usage credits." Credits must be purchased prior to using the API, and your credits will be applied to your usage according to our current **[pricing](https://claude.com/pricing#api)**. These credits can be used for API access, Workbench usage, and Claude Code. Failed requests are not charged, and you will only be billed for successful API calls and completed tasks.

You can track credit usage in your **[Claude Console settings on the Billing page](https://platform.claude.com/settings/billing)**. If you run out of credits, you will no longer be able to call the API or use Workbench. You can choose to set up an auto-reload that purchases additional credits when your balance falls below a set limit.

To add usage credits and adjust your auto-reload settings, navigate to your Billing page. Click on the “Buy credits” button, and enter in the amount of credits you would like to purchase. After you’ve purchased your credits, they will be immediately available to you.

You can also view your organization's available credit balance in your **[Claude Console settings on the Billing page](https://platform.claude.com/settings/billing)**.

To adjust your auto-reload options, click on “Edit” in the auto-reload section. You can toggle auto-reload on or off here. If choosing to use auto-reload, you can set the minimum account balance, and the amount to reload to if your account reaches that minimum balance.

Please note that purchased credits are subject to our **[Credit Terms](https://www.anthropic.com/legal/credit-terms)**. Credits expire one year from the purchase date, and this expiration date cannot be extended. Expired credits will appear in your **[Console billing settings](https://platform.claude.com/settings/billing)** under **Invoice history**. All credit purchases are non-refundable.

Learn more about credit usage and how it relates to usage limits in our **[Claude API docs](../04-API-Reference/Other/api-rate-limits-28d3302987.md)**.

## How to update your Console payment method

1. Log in to your Console account with an Admin or Billing role.

2. Navigate to **[Settings \> Billing](https://platform.claude.com/settings/billing)**.

3. Click the pencil icon next to your current payment method to update your card information.

4. Enter your new card details in the **Update payment method** modal, then click "Update."

---
## how-do-i-pay-for-my-claude-api-usage-0b490bdd29.md

# How do I pay for my Claude API usage?

August 19, 2026

Copy for LLM

## Prepaid usage credits

You're billed only for successful API calls and completed tasks. Failed requests aren't charged.

**Note:** If your client disconnects or times out in the middle of a request that was on track to succeed, that request is still charged.

If you run out of credits, you can no longer call the API or use the playground until you add more.

## Buy credits

1. Log in to the Console with an Admin or Billing role.

2. Navigate to **[Settings \> Billing](https://platform.claude.com/settings/billing)**.

3. Click "Buy credits."

4. Enter the amount of credits you want to purchase and confirm.

Purchased credits are available immediately. Your organization's available credit balance and credit usage are shown on the same **Billing** page.

## Set up auto-reload

Auto-reload purchases additional credits automatically when your balance falls below a threshold you set.

1. On the **Billing** page, click "Edit" in the **Auto-reload** section.

2. Toggle auto-reload on or off.

3. If auto-reload is on, set the minimum balance that triggers a purchase and the amount to reload to.

## Credit expiration and refunds

Learn more about how credit usage relates to rate limits in the **[Claude API docs](https://platform.claude.com/docs/en/api/rate-limits)**.

## Monthly invoicing

Some organizations are billed monthly in arrears instead of buying credits upfront. This applies to Console organizations with an invoicing arrangement set up through our Sales team.

You can view charges for the current billing period on the **[Billing](https://platform.claude.com/settings/billing)** page in your Console settings.

If you need custom rate limits, monthly invoicing, or hands-on support, **[contact our Sales team](https://claude.com/contact-sales)**.

## Update your Console payment method

1. Log in to the Console with an Admin or Billing role.

2. Navigate to **[Settings \> Billing](https://platform.claude.com/settings/billing)**.

3. Click the pencil icon next to your current payment method.

4. Enter your new card details in the **Update payment method** modal, then click "Update."

---
## how-do-usage-and-length-limits-work-1a386ba361.md

# How do usage and length limits work?

Copy for LLM

When working with Claude, you may encounter two different types of limits that work in distinct ways: **usage limits** and **length limits**. Understanding the difference between these can help you use Claude more effectively.

## What are usage limits?

Usage limits control how much you can interact with Claude over a specific time period. Think of this as your "conversation budget" that determines how many messages you can send to Claude, or how long you can work with Claude Code, before needing to wait for your limit to reset.

Your usage is affected by several factors, including the length and complexity of your conversations, the features you use, which Claude model you're chatting with, and the effort level you've selected. Different subscription plans (Pro, Max, Team, etc.) have different usage allowances, with paid plans offering higher limits.

Note that your usage of all different Claude product surfaces (claude.ai, Claude Code, Claude Desktop) counts towards the same usage limit.

Learn more about **[changing the model, effort, and thinking settings](https://support.claude.com/en/articles/8664678)**.

## How can I get unlimited usage?

There are a couple of different ways to increase your usage depending on your plan:

- If you're using a paid plan, including Pro, Max, Team, or seat-based Enterprise plans, see these articles for details about purchasing usage credits:

- **[Manage usage credits for paid Claude plans](https://support.claude.com/en/articles/12429409-)**

- **[Manage usage credits for Team and seat-based Enterprise plans](https://support.claude.com/en/articles/12005970-)**

- If your organization has a usage-based Enterprise plan, your usage is based on consumption. See this article for additional information: **[How am I billed for my Enterprise plan?](https://support.claude.com/en/articles/11526368-how-am-i-billed-for-my-enterprise-plan)**

For strategies to maximize your message allotment, see **[Usage limit best practices](https://support.claude.com/en/articles/9797557-usage-limit-best-practices)**.

## What are length limits?

Length limits relate to Claude's context window—the amount of information Claude can work with in a single chat. Think of the context window as Claude's working memory that determines how much content it can process and remember at once.

**Note**: The context window size depends on which model you're using. On paid plans, the newest models support up to a 1M token context window, while others support 500K or 200K tokens. A portion of the context window is always reserved for Claude's response, so the longest possible conversation is slightly smaller than the full window. For the current per-model breakdown, refer to **[How large is the context window on paid Claude plans?](https://support.claude.com/en/articles/8606394-how-large-is-the-context-window-on-paid-claude-plans)**

## Automatic context management

For users with code execution enabled, Claude now automatically manages long conversations. When your conversation approaches the context window limit, Claude summarizes earlier messages to continue the conversation seamlessly. This means you can have longer, more natural conversations with fewer interruptions.

Your full chat history is preserved so Claude can reference it even after summarization. You may occasionally see that Claude is "organizing its thoughts" during long conversations—this indicates automatic context management is working.

Longer conversations that trigger automatic context management consume more of your usage limit. Try starting a new conversation if you're approaching your usage limit in a longer chat.

**Note:** **[Code execution must be enabled](https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude#h_1c99382190)** for automatic context management. Rare edge cases (such as very large first messages) may still encounter context limits.

## How can I increase the size of Claude’s context window?

While you can't increase the fixed context window size for your plan, you can use these strategies to maximize available context space and optimize both your context window and usage limits:

- **Utilize projects effectively:** Projects use retrieval-augmented generation (RAG), which allows Claude to work with larger amounts of information more efficiently by only loading relevant content into the context window.

- **Shorten project instructions:** Keep your project instructions concise and focused on essential information. Claude performs best when you use project instructions for general context around your project, key guidelines, and Claude's role. Reserve task-specific instructions for the chat itself.

- **Remove unused project files:** Regularly clean up files you're no longer actively using in your projects.

- **Toggle extended thinking off:** Turn off this feature when you don't need Claude's enhanced reasoning for a particular task.

- **Lower the effort level:** Choose a lower effort level for routine tasks that don't need Claude's most thorough responses. Higher effort uses more tokens.

- **Turn off tools you don't need:** Turn off apps you've connected when a conversation doesn't need them, and ask Claude not to search the web when you don't need current information.

**Note:** Tools and connectors are token-intensive, so managing them helps both maximize your available context window and optimize your usage limits.

## Key differences

The main distinction is that usage limits control *how much* you can use Claude across all your conversations, while length limits control *how long* any single conversation can become. Usage limits are about quantity over time, while length limits are about the depth and complexity of individual conversations.

If you hit your usage limit, you'll need to wait for it to reset, upgrade your plan, or purchase usage credits. If you hit a length limit, you can start a new conversation or use **[features like projects](https://support.claude.com/en/articles/9517075-what-are-projects)** to work with larger amounts of information more efficiently.

---
## how-do-usage-and-length-limits-work.md

# How do usage and length limits work?

When chatting with Claude, you may encounter two different types of limits that work in distinct ways: **usage limits** and **length limits**. Understanding the difference between these can help you use Claude more effectively.

## What are usage limits?

Your usage is affected by several factors, including the length and complexity of your conversations, the features you use, and which Claude model you're chatting with. Different subscription plans (Pro, Max, Team, etc.) have different usage allowances, with paid plans offering higher limits.

Note that your usage of all different Claude product surfaces (claude.ai, Claude Code, Claude Desktop) counts towards the same usage limit.

## How can I get unlimited usage?

There are a couple of different ways to increase your usage depending on your plan:

- If you’re using a paid plan, including Pro, Max, Team, or seat-based Enterprise plans, see these articles for details about purchasing extra usage:

- **[Extra usage for paid Claude plans](../17-Billing-Plans/extra-usage-for-paid-claude-plans.md)**

- **[Extra usage for Team and seat-based Enterprise plans](../17-Billing-Plans/extra-usage-for-team-and-seat-based-enterprise-plans.md)**

- If your organization has a usage-based Enterprise plan, your usage is based on consumption. See this article for additional information: **[How am I billed for my Enterprise plan?](../17-Billing-Plans/how-am-i-billed-for-my-enterprise-plan.md)**

For strategies to maximize your message allotment, see **[Usage limit best practices](https://support.claude.com/en/articles/9797557-usage-limit-best-practices)**.

## What are length limits?

Claude's context window size is 200K tokens across all models and paid plans, except for Enterprise plans, which have a 500K context window on some models. For more information, refer to **[What is the Enterprise plan?](../17-Billing-Plans/what-is-the-enterprise-plan.md)**

## Automatic context management

## How can I increase the size of Claude’s context window?

While you can't increase the fixed context window size for your plan, you can use these strategies to maximize available context space and optimize both your context window and usage limits:

- **Remove unused project files:** Regularly clean up files you're no longer actively using in your projects.

- **Toggle extended thinking off:** Turn off this feature when you don't need Claude's enhanced reasoning for a particular task.

- **Temporarily disable non-critical tools and connectors:** Disable web search, Research, and MCP connectors from your "Search and tools" settings when they're not needed for specific conversations.

**Note:** Tools and connectors are token-intensive, so managing them helps both maximize your available context window and optimize your usage limits.

## Key differences

If you hit your usage limit, you'll need to wait for it to reset, upgrade your plan, or purchase extra usage. If you hit a length limit, you can start a new conversation or use **[features like projects](https://support.claude.com/en/articles/9517075-what-are-projects)** to work with larger amounts of information more efficiently.

---
## how-to-use-claude-in-your-preferred-language-32b39d6142.md

# How to use Claude in your preferred language

August 6, 2026

Copy for LLM

Claude is available in multiple languages on web and desktop applications.

## Supported Languages

- English

- French

- German

- Hindi

- Indonesian

- Italian

- Japanese

- Korean

- Portuguese (Brazilian)

- Spanish (Latin America)

- Spanish (Spain)

## Changing your language setting

### On Web and Desktop

1. Click your profile icon in the lower left corner.

2. Go to the "Language" section.

3. Choose your preferred language.

4. The interface will update automatically.

**Note:** Even if you've changed your language settings, you can still chat with Claude in any language. Claude will converse with you in the language you use.

## Voice mode languages

Voice mode has its own language setting, separate from your display language. To change the language Claude speaks and listens to in voice mode, go to **[Settings \> General](https://claude.ai/settings/general)**, then find **Voice \> Language** and choose a language. Learn more in **[Use voice mode](https://support.claude.com/en/articles/11101966-use-voice-mode)**.

---
## how-to-use-claude-in-your-preferred-language.md

4. How to use Claude in your preferred language

# How to use Claude in your preferred language

Claude is available in multiple languages on web and desktop applications.

## Supported Languages

- English

- French

- German

- Hindi

- Indonesian

- Italian

- Japanese

- Korean

- Portuguese (Brazilian)

- Spanish (Latin America)

- Spanish (Spain)

## Changing your language setting

### On Web and Desktop

1. Click your profile icon in the lower left corner.

2. Go to the "Language" section.

3. Choose your preferred language.

4. The interface will update automatically.

**Note:** Even if you've changed your language settings, you can still chat with Claude in any language. Claude will converse with you in the language you use.

---
## i-ve-been-using-other-large-language-models-llms-to-power-my-product-how-should-a86d0af4e1.md

# I’ve been using other large language models (LLMs) to power my product. How should I alter my prompts when switching to Claude?

March 16, 2026

Copy for LLM

You can use the prompt improver in the Claude Console to automatically adapt prompts that were originally written for other AI models.

---
## i-ve-been-using-other-large-language-models-llms-to-power-my-product-how-should.md

4. I’ve been using other large language models (LLMs) to power my product. How should I alter my prompts when switching to Claude?

# I’ve been using other large language models (LLMs) to power my product. How should I alter my prompts when switching to Claude?

You can use the prompt improver in the Claude Console to automatically adapt prompts that were originally written for other AI models.

---
## law-enforcement-requests-2e51083860.md

# Law Enforcement Requests

March 16, 2026

Copy for LLM

Anthropic PBC discloses account records solely in accordance with our Terms of Service and applicable law. For more information about Anthropic’s policies for handling governmental requests for user information, see [here](https://support.anthropic.com/en/articles/9519291-what-is-anthropic-s-policy-for-handling-governmental-requests-for-user-information). If you are a law enforcement agent who is authorized to gather evidence in connection with an official investigation, you may request records from Anthropic by emailing [\[email protected\]](/cdn-cgi/l/email-protection#c5a9a4b2e8a0aba3aab7a6a0a8a0abb1e8b7a0b4b0a0b6b1b685a4abb1adb7aab5aca6eba6aaa8).

Please include the following information with your request:

- Law Enforcement Agency Country

- City

- State/Province

- Zip/Postal Code

- Law Enforcement Address

- Name of Court/Policy Department/Authority/Agency

- Name of Contact Person Handling this Matter (Detective, Investigator, etc.)

- Official Law Enforcement Agency Request Email Address

- Contact Person’s Phone Number

- Description of legal Inquiry

- Matter Type of legal Inquiry

- Attach Any Relevant Documentation (e.g., Subpoena, Warrant)

- NCMEC Report Identifier (If Applicable)

---
## law-enforcement-requests.md

# Law Enforcement Requests

Anthropic PBC discloses account records solely in accordance with our Terms of Service and applicable law. For more information about Anthropic’s policies for handling governmental requests for user information, see [here](https://support.anthropic.com/en/articles/9519291-what-is-anthropic-s-policy-for-handling-governmental-requests-for-user-information). If you are a law enforcement agent who is authorized to gather evidence in connection with an official investigation, you may request records from Anthropic by emailing [\[email protected\]](/cdn-cgi/l/email-protection#c7aba6b0eaa2a9a1a8b5a4a2aaa2a9b3eab5a2b6b2a2b4b3b487a6a9b3afb5a8b7aea4e9a4a8aa).

Please include the following information with your request:

- Law Enforcement Agency Country

- City

- State/Province

- Zip/Postal Code

- Law Enforcement Address

- Name of Court/Policy Department/Authority/Agency

- Name of Contact Person Handling this Matter (Detective, Investigator, etc.)

- Official Law Enforcement Agency Request Email Address

- Contact Person’s Phone Number

- Description of legal Inquiry

- Matter Type of legal Inquiry

- Attach Any Relevant Documentation (e.g., Subpoena, Warrant)

- NCMEC Report Identifier (If Applicable)

---
## legal-and-compliance-claude-code-docs-2d64f2bee3.md

# Legal and compliance

Legal agreements, compliance certifications, and security information for Claude Code.

> ## Documentation Index
>
> Fetch the complete documentation index at: <https://code.claude.com/docs/llms.txt>
>
> Use this file to discover all available pages before exploring further.

Starting June 15, 2026, Agent SDK and `claude -p` usage on subscription plans will draw from a new monthly Agent SDK credit, separate from your interactive usage limits. See [Use the Claude Agent SDK with your Claude plan](https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan) for details.

Legal agreements

License

Your use of Claude Code is subject to:

- [Commercial Terms](https://www.anthropic.com/legal/commercial-terms) - for Team, Enterprise, and Claude API users
- [Consumer Terms of Service](https://www.anthropic.com/legal/consumer-terms) - for Free, Pro, and Max users

Commercial agreements

Whether you’re using the Claude API directly (1P) or accessing it through Amazon Bedrock or Google Vertex (3P), your existing commercial agreement will apply to Claude Code usage, unless we’ve mutually agreed otherwise.

Compliance

Healthcare compliance (BAA)

If a customer has a Business Associate Agreement (BAA) with us, and wants to use Claude Code, the BAA will automatically extend to cover Claude Code if the customer has executed a BAA and has [Zero Data Retention (ZDR)](/docs/en/zero-data-retention) activated. The BAA will be applicable to that customer’s API traffic flowing through Claude Code. ZDR is enabled on a per-organization basis, so each organization must have ZDR enabled separately to be covered under the BAA.

Usage policy

Acceptable use

Claude Code usage is subject to the [Anthropic Usage Policy](https://www.anthropic.com/legal/aup). Advertised usage limits for Pro and Max plans assume ordinary, individual usage of Claude Code and the Agent SDK.

Authentication and credential use

Claude Code authenticates with Anthropic’s servers using OAuth tokens or API keys. These authentication methods serve different purposes:

- **OAuth authentication** is intended exclusively for purchasers of Claude Free, Pro, Max, Team, and Enterprise subscription plans and is designed to support ordinary use of Claude Code and other native Anthropic applications. More information about how users can authenticate with OAuth tokens can be found in [Logging in to your Claude account](../21-Account-Support/logging-in-to-your-claude-account.md).
- **Developers** building products or services that interact with Claude’s capabilities, including those using the [Agent SDK](/docs/en/agent-sdk/overview), should use API key authentication through [Claude Console](../04-API-Reference/Other/platform-claude-com.md) or a supported cloud provider. Anthropic does not permit third-party developers to offer Claude.ai login or to route requests through Free, Pro, or Max plan credentials on behalf of their users.

Anthropic reserves the right to take measures to enforce these restrictions and may do so without prior notice. For questions about permitted authentication methods for your use case, please [contact sales](https://www.anthropic.com/contact-sales?utm_source=claude_code&utm_medium=docs&utm_content=legal_compliance_contact_sales).

Security and trust

Trust and safety

You can find more information in the [Anthropic Trust Center](https://trust.anthropic.com) and [Transparency Hub](https://www.anthropic.com/transparency).

Security vulnerability reporting

Anthropic manages our security program through HackerOne. [Use this form to report vulnerabilities](https://hackerone.com/4f1f16ba-10d3-4d09-9ecc-c721aad90f24/embedded_submissions/new).

------------------------------------------------------------------------

---
## manage-api-key-environment-variables-in-claude-code-b2ab15dd01.md

# Manage API key environment variables in Claude Code

May 5, 2026

Copy for LLM

When using Claude Code, it's important to understand how authentication methods are prioritized to avoid unexpected API charges and ensure you're using your intended account.

## How authentication works

- Claude Code prioritizes environment variable API keys over authenticated subscriptions.

- This is intentional behavior designed to give you flexibility in choosing your authentication method.

- When an API key is set as an environment variable, you'll be charged via API pay-as-you-go rates using the API account associated with that key.

- This happens even if you're logged into Claude Code with a claude.ai subscription or a different Console account.

## Best practices

**To use Claude Code with your Claude subscription:** Keep the ANTHROPIC_API_KEY environment variable unset.

- This prevents unexpected API charges and ensures you're using your subscription's included usage.

- If you need to use a specific API key occasionally, set it temporarily only when needed.

- Run /status in Claude Code periodically to verify your current authentication method.

## Authentication conflict warnings

Claude Code will notify you when there's a conflict between your authenticated subscription and an environment variable API key:

1. During initial setup, if an API key is detected in your environment variables, Claude Code will ask you to confirm which authentication method you want to use.

2. After successful login, you'll see a notification if both credentials are active, alerting you to the potential for unexpected API charges.

## Check your current configuration

To verify if an API key is set as an environment variable, run /status in Claude Code. This will show you which authentication method is currently active.

To check your environment variable directly, run one of these commands in a terminal (outside of Claude Code):

macOS/Linux:

echo $ANTHROPIC_API_KEY

Windows CMD:

echo %ANTHROPIC_API_KEY%

Windows PowerShell:

echo $env:ANTHROPIC_API_KEY

## Set an API key temporarily

If you need to use an API key for the current terminal session only:

macOS/Linux:

export ANTHROPIC_API_KEY='your-api-key-here'

Windows CMD:

set ANTHROPIC_API_KEY=your-api-key-here

Windows PowerShell:

$env:ANTHROPIC_API_KEY="your-api-key-here"

## Set an API key environment variable permanently

macOS/Linux:

For zsh (default on macOS):
bash
# Add to shell config file
echo 'export ANTHROPIC_API_KEY="your-api-key-here"' >> ~/.zshrc

# Apply changes
source ~/.zshrc
For bash:
bash
# Add to shell config file
echo 'export ANTHROPIC_API_KEY="your-api-key-here"' >> ~/.bash_profile

# Apply changes
source ~/.bash_profile

Windows:

1. Open System Properties → Advanced → Environment Variables

2. Under "User variables", click "New"

3. Variable name: ANTHROPIC_API_KEY

4. Variable value: your-api-key-here

5. Click OK and restart your terminal

## Remove an API key environment variable

macOS/Linux (temporary):

unset ANTHROPIC_API_KEY

macOS (permanent):

# Remove from config file
sed -i '' '/ANTHROPIC_API_KEY/d' ~/.zshrc
source ~/.zshrc

Linux (permanent)

sed -i '/ANTHROPIC_API_KEY/d' ~/.zshrc

Windows CMD:

set ANTHROPIC_API_KEY=

Windows PowerShell:

Remove-Item Env:ANTHROPIC_API_KEY

Windows (permanent): Delete the variable from System Environment Variables settings.

## Common issues to avoid

- Setting environment variables in shell configuration files and forgetting about them.

- Not restarting your terminal after changing environment variables.

- Assuming you're using your subscription when an API key is configured in your environment.

If you have any questions, please **[contact our Support team](https://support.claude.com/en/articles/9015913-how-to-get-support)**.

---
## manage-claude-cowork-plugins-for-your-organization-09f430a953.md

# Manage plugins for your organization

Copy for LLM

Plugin marketplaces let Team and Enterprise plan owners distribute curated plugins to everyone in their organization. You create a marketplace, add plugins to it, and control exactly which plugins your team members can see and use. Plugins you distribute appear in both chat (on the web and the Chat tab in Claude Desktop) and Claude Cowork.

Owners and Primary Owners of Team and Enterprise plans can manage organization plugins in **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**.

**Requirements:** Cowork and Skills must both be enabled for your organization before you can use plugin marketplaces.

------------------------------------------------------------------------

## Use Anthropic-built marketplaces

Anthropic provides built-in marketplaces of plugins for different functions, such as legal and finance. A Knowledge Work marketplace is added to your organization by default.

To add an Anthropic-built marketplace:

1. Go to **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**.

2. Click "Add plugins."

3. Select "Browse Anthropic sources."

4. Click "Add" on each one you want to appear for everyone in your organization.

To remove the Knowledge Work marketplace if it isn't relevant to your teams:

1. Go to **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**.

2. Find **Knowledge Work**.

3. Click the menu button in the upper right corner.

4. Select "Remove."

Anthropic-built marketplaces show readable names, like "Knowledge Work," instead of the underlying repository slug.

------------------------------------------------------------------------

## Create your own marketplace

Beyond the Anthropic-built marketplaces, you can create your own. There are two ways to add plugins to a marketplace you create:

- **Manual upload:** Upload individual plugin ZIP files through the admin UI. Best for quick iteration, one-off tools, or teams that don't use GitHub for plugin development.

- **GitHub syncing:** Connect a private GitHub repository and Cowork automatically syncs plugins from it. Best when multiple developers collaborate on plugins or you want version-controlled updates.

You can use both approaches in parallel. For example, run a GitHub-synced marketplace for your core plugins and a separate manual marketplace for ad-hoc tools.

### Set up a manual marketplace

1. Go to **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**.

2. Click "Add plugins" and select "Upload a file" as the source.

3. If this is your first time setting up a marketplace, "Upload to a new marketplace" is auto-selected. You'll be able to choose "Add to an existing marketplace" when uploading plugins later.

4. Enter a name for your marketplace.

5. Either drag your files in, or click the upload prompt and select your file. The file must be a valid .zip under 50 MB.

6. Repeat for each plugin you want to add.

7. Click "Upload" to add your plugins to a new marketplace.

If you upload a plugin with the same name as an existing one, it overwrites the previous version automatically. You don't need to delete the old one first.

### Set up a GitHub-synced marketplace

**Prepare your repository**

Your repository must be **private or internal**—public repos aren't allowed for organization marketplaces. You can connect a repo hosted on github.com or on a self-hosted GitHub Enterprise Server instance.

For details on plugin structure and formatting, see the **[plugin reference documentation](https://code.claude.com/docs/en/plugins-reference)**.

Additional resources:

- **[Create and distribute a plugin marketplace](https://code.claude.com/docs/en/plugin-marketplaces)**

- **[Create plugins](https://code.claude.com/docs/en/plugins)**

- **[Creating a new GitHub repository](https://docs.github.com/en/repositories/creating-and-managing-repositories/creating-a-new-repository)**

**Connect the repository**

1. Make sure both Cowork and Skills are enabled for your organization.

1. **[Enable Cowork](https://support.claude.com/en/articles/13455879-use-cowork-on-team-and-enterprise-plans#h_71cdc52dfc)**

2. **[Enable Skills](https://support.claude.com/en/articles/13119606-provision-and-manage-skills-for-your-organization#h_7673241237)**

2. Go to **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**.

3. Click "Add plugin" and select "GitHub" as the source.

4. Enter the repository in `owner/repo` format (for example, `acme-corp/claude-plugins`).

Your personal GitHub token is verified to confirm you have access, then Cowork uses its GitHub App installation token for sync operations.

**Can't see your repo?** Make sure the Claude GitHub App is installed in that repository.

**How syncing works**

An initial sync runs automatically when you connect a repository. After that, organization owners can opt-in to continued automatic updates per marketplace by going to **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**, clicking the menu button in the upper right corner of the marketplace, then toggling "Sync automatically" on:

Enabling automatic sync creates a webhook on the connected repository. The person turning the toggle on must have admin-level access to that repository on GitHub. This is checked through their personal GitHub connection, which is separate from the Claude GitHub App installation. Without admin access, the page shows "Cannot access repository. Ensure the repository exists and the Claude GitHub App is installed," even when the App is installed correctly and manual updates work.

The Claude GitHub App's **Webhooks (Read & Write)** permission must also be approved on the installation. On installations created before this permission was introduced, GitHub shows a "Claude is requesting updated permissions" prompt that a repository or organization admin needs to approve.

Once enabled, automatic sync runs when a pull request that includes a plugin version bump is merged to the repository's default branch. Direct pushes to the default branch don't trigger a sync. You can always trigger a sync manually by clicking "Update" on the marketplace.

During a sync, Cowork compares the latest commit in your repo against the last-synced commit. If nothing has changed, the sync is skipped. If there are changes, Cowork reads the manifest, validates each plugin, and replaces all plugins in the marketplace with the current state of the repo. Syncs can take up to 30 minutes depending on the number of plugins.

**Important:** If a sync fails, plugins may be temporarily removed for your team members. If this happens, check the failure message, fix the issue in your repo, push the update, and trigger the sync again. Once the sync succeeds, verify that your installation preferences are still set correctly, since they may have been reset during the failure.

### Choose between manual upload and GitHub sync

[TABLE]

------------------------------------------------------------------------

## Control plugin distribution

Once your marketplace has plugins, you control how they're distributed using installation preferences. For each plugin, you can set one of four options:

[TABLE]

### Set preferences

1. In **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**, navigate to your marketplace.

2. Select the installation preference for each plugin.

3. Changes take effect on each member's next session or plugin refresh.

### What members experience

Members browse available plugins through the **Browse plugins** modal. Auto-installed plugins appear in their installed list automatically. Available plugins show up in the catalog for self-service installation.

Members can't edit organization-managed plugins, which prevents conflicting changes to shared tooling.

------------------------------------------------------------------------

## Customize plugin access by group

Enterprise admins can override a plugin's organization-wide installation preference for specific groups. For example, you can auto-install a plugin for the Engineering group, make it available for Legal to install on their own, and hide it from everyone else.

Group-level plugin access is available on Enterprise plans and configurable by Admins and above.

### How group overrides work

Each plugin in your marketplace has an organization-wide installation preference (Installed by default, Available for install, Required, or Not available). By default, every group inherits that organization-wide setting.

When you set a group-level override for a plugin, it replaces the org-wide setting for members of that group. The resolution order is: group setting, then org-wide setting, then marketplace default.

### Set plugin access for a group

1. In **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**, navigate to your marketplace.

2. Find the plugin you want to customize.

3. In the **Custom access** column, click “Add groups.”

4. Select the group and choose one of the installation preferences listed above.

Both manually created groups and SCIM-provisioned groups from your identity provider appear in the group picker and work the same way.

### What happens when a member is in multiple groups

If a member belongs to two or more groups with different settings for the same plugin, the **most permissive** setting applies. The order from most to least permissive is: Required \> Installed by default \> Available for install \> Not available.

For example, if Group A sets a plugin to "Not available" and Group B sets it to "Installed by default," a member in both groups gets the plugin installed by default.

**Note:** This differs from how group spend limits resolve. Spend limits apply either the higher or lower group value depending on your **Multi-group spend limit** setting. Plugin access applies the most permissive value, because groups here are meant to enable access for teams that need a tool, not to act as a security boundary. If you need to hard-block a plugin, set its org-wide preference to "Not available" and only grant access to the groups that should have it.

### What happens when a group is deleted

If a group is removed (for example, deleted from your identity provider), the override remains in the admin UI but is flagged as orphaned. It has no effect on members (since no one belongs to a deleted group) and doesn't count toward the custom access badge. You can clear orphaned overrides from the plugin's custom access settings.

### Do group settings persist across marketplace re-syncs?

Yes. Group-level overrides persist when you re-sync a GitHub-connected marketplace. They're only removed if the plugin itself is deleted from the marketplace.

------------------------------------------------------------------------

## Update and remove plugins

### Manual marketplaces

To update a plugin, upload a new ZIP file with the same plugin name. The new version overwrites the existing one automatically. Plugin names are the unique identifier, so `legal` will always replace `legal`.

To remove a plugin, delete it from your marketplace in **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**.

### GitHub-synced marketplaces

Push your changes to the connected repository, then go to **[Organization settings \> Plugins](https://claude.ai/admin-settings/plugins)**, find your marketplace, and click "Update" to trigger a sync. Each sync replaces all plugins with the current state of the repo. If an owner has enabled "Sync automatically" for the marketplace, a sync also runs whenever a pull request with a plugin version bump is merged to the repository's default branch.

To remove a plugin, delete it from the repository and trigger a sync.

------------------------------------------------------------------------

## Limits

[TABLE]

------------------------------------------------------------------------

## Naming rules

Plugin names must use **lowercase words separated by hyphens** (for example, `deployment-tools`, not `Deployment Tools`). The following marketplace names are reserved and can't be used:

- `claude-code-marketplace`

- `claude-code-plugins`

- `claude-plugins-official`

- `anthropic-marketplace`

- `anthropic-plugins`

- `agent-skills`

- `life-sciences`

Names that impersonate official Anthropic marketplaces are also blocked.

**Note:** Plugin authors can set a `displayName` so a plugin shows a readable name in the catalog instead of its slug. If an author doesn't set one, the plugin shows its slug (for example, `deployment-tools`).

------------------------------------------------------------------------

## Troubleshooting

### Upload rejected

Common causes: the file exceeds 50 MB, it isn't a valid ZIP file, or the marketplace has reached the 100-plugin limit. Check the file size and format, and remove unused plugins if you're at capacity. If skill and plugin scanning is on, a plugin can also be rejected because it was flagged for malicious content. Review the reason, fix the issue, and upload the plugin again.

### Plugin not appearing for members

Check the plugin's installation preference in your marketplace settings. If it's set to **Not available**, members won't see it. Also confirm that Cowork and Skills are both enabled for your organization.

### Updated plugin not reflecting for members

Changes take effect on each member's next session or plugin refresh. If the update still isn't showing, confirm the upload succeeded by checking the plugin version in your marketplace.

### GitHub sync fails with a content error

One or more plugins in your repo is likely formatted incorrectly. Fix the formatting issue, push the update to GitHub, and trigger the sync again. For plugin structure requirements, see the **[plugin reference documentation](https://code.claude.com/docs/en/plugins-reference)**.

### Sync fails with "External plugin sources are not yet supported," or plugins are skipped with "Repository not found on github.com. Check the URL and make sure the repository is public."

One or more plugin entries in your `marketplace.json` use a `source` that points outside the connected repository (a `github`, `url`, or `git-subdir` source), and the target repository is private. The organization sync can currently only fetch external sources from public repositories. Move the plugin folders into the marketplace repository and change each entry's `source` to a relative path (for example, `"./plugins/my-plugin"`), then push and re-sync. Alternatively, upload the affected plugins individually via **Customize \> Add plugin \> Create plugin \> Upload plugin**.

### Plugins disappeared after a failed sync

A failed GitHub sync can temporarily remove plugins from your marketplace. Fix the underlying issue, re-sync successfully, then verify that installation preferences are set correctly, since they may have been reset.

### Can't see a GitHub repo when connecting

Make sure the Claude GitHub App is installed in that repository. Your personal GitHub token is checked first to confirm access, but the sync itself uses the GitHub App installation token.

### "Cannot access repository" when turning on "Sync automatically"

If manual updates work but turning on "Sync automatically" shows "Cannot access repository. Ensure the repository exists and the Claude GitHub App is installed," there are two likely causes:

- **You don't have admin access to the repository.** Turning on automatic sync creates a webhook, which requires admin-level access to the repo through your personal GitHub connection. Ask a repository admin to enable the toggle, or have your GitHub access upgraded.

- **The Claude GitHub App's Webhooks permission hasn't been approved.** On older installations, GitHub shows a "Claude is requesting updated permissions" prompt. A repository or organization admin needs to approve the **Webhooks (Read & Write)** permission on the installation.

---
## manage-claude-s-tool-access-0ef5e7a423.md

# Manage Claude’s tool access

March 16, 2026

Copy for LLM

When you connect many services to Claude, you can control how those connectors are loaded into your conversations. This helps Claude work more accurately and efficiently, especially if you've added 10 or more connectors.

Tool access modes are available for all users on Claude, Cowork, Claude Desktop, and Claude Mobile (iOS and Android).

## Why tool access matters

Each connector takes up space in a conversation. With a small number of connected services, this is rarely noticeable. But as your connector library grows, the combined overhead can limit how much room is left for your actual work: documents, code, and conversation history.

To address this, Claude supports three tool access modes that control when and how your connectors are loaded.

------------------------------------------------------------------------

## Tool access modes

You can manage Claude’s access to your connectors per conversation using the **Tool access** setting in your chat.

Choose from three options:

- **Auto** *(default)*: Claude decides dynamically which connectors to load based on what you're working on. This is the best starting point for most users.

- **Always available**: All your connectors are loaded at the start of every conversation. Claude can use any of them immediately, without any extra steps.

- Best for: Fewer than 10 connectors you use constantly.

- Trade-off: Uses more conversation space upfront.

- **On demand:** Connectors aren't loaded until Claude searches for the right one based on your request. Claude finds the most relevant connectors and loads only those.

- Best for: Large connector libraries (10 or more), or when you're running into conversation length issues.

- Trade-off: Claude may take an extra step to find the right connector before using it.

------------------------------------------------------------------------

## Which mode should I use?

[TABLE]

------------------------------------------------------------------------

## How to change your tool access setting

You can set your tool access mode per conversation by following these steps:

1. Open a chat with Claude.

2. Click the “+” button in the lower left corner of your chat, or type "/" to open the menu.

3. Mouse over “Connectors,” then “Tool access.”

4. Select your preferred mode from the three options.

Your selection only applies to that conversation. You can change it at any time.

------------------------------------------------------------------------

## Frequently asked questions

### Will “On demand” mode miss the connectors I need?

Claude searches for the most relevant connectors based on your request, so it works well for most tasks. For connectors that need to be available every time, use “Always available” or set that connector's access individually.

### Why does Claude sometimes add an extra message to use a connector?

In “On demand” mode, Claude searches for the right tool before using it. This search step adds one interaction, but keeps your conversation from hitting length limits when you have many tools connected.

---
## manage-claude-s-tool-access.md

# Manage Claude’s tool access

Tool access modes are available for all users on Claude, Cowork, Claude Desktop, and Claude Mobile (iOS and Android).

## Why tool access matters

To address this, Claude supports three tool access modes that control when and how your connectors are loaded.

------------------------------------------------------------------------

## Tool access modes

You can manage Claude’s access to your connectors per conversation using the **Tool access** setting in your chat.

Choose from three options:

- **Auto** *(default)*: Claude decides dynamically which connectors to load based on what you're working on. This is the best starting point for most users.

- **Always available**: All your connectors are loaded at the start of every conversation. Claude can use any of them immediately, without any extra steps.

- Best for: Fewer than 10 connectors you use constantly.

- Trade-off: Uses more conversation space upfront.

- **On demand:** Connectors aren't loaded until Claude searches for the right one based on your request. Claude finds the most relevant connectors and loads only those.

- Best for: Large connector libraries (10 or more), or when you're running into conversation length issues.

- Trade-off: Claude may take an extra step to find the right connector before using it.

------------------------------------------------------------------------

## Which mode should I use?

[TABLE]

------------------------------------------------------------------------

## How to change your tool access setting

You can set your tool access mode per conversation by following these steps:

1. Open a chat with Claude.

2. Click the “+” button in the lower left corner of your chat, or type "/" to open the menu.

3. Mouse over “Connectors,” then “Tool access.”

4. Select your preferred mode from the three options.

Your selection only applies to that conversation. You can change it at any time.

------------------------------------------------------------------------

## Frequently asked questions

### Will “On demand” mode miss the connectors I need?

### Why does Claude sometimes add an extra message to use a connector?

---
## manage-cowork-plugins-for-your-organization-98f0dc9ff0.md

# Manage plugins for your organization

Copy for LLM

Plugin marketplaces let Team and Enterprise plan owners distribute curated plugins to everyone in their organization. You create a marketplace, add plugins to it, and control exactly which plugins your team members can see and use. Plugins you distribute appear in chat (on the web and the Chat tab in Claude Desktop), in Claude Cowork, and in Claude Code sessions where members sign in with the same Claude account.

Owners and Primary Owners of Team and Enterprise plans can manage organization plugins in **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**. On Enterprise plans, members with a custom role that includes managing the organization's libraries can too.

**Requirements:** Cowork and Skills must both be enabled for your organization before you can use plugin marketplaces.

For a reference version of these controls, see **[Manage plugins for your organization](https://code.claude.com/docs/en/plugins/org)** in the Claude Code docs.

**Note:** Turning off Skills for your organization also stops skills and plugins from syncing to Claude Code, and removes the ones that already synced there. Turning off Cowork doesn't affect the sync to Claude Code. To keep skills and plugins in Claude but stop only the sync, set `syncClaudeAiSkills` and `syncClaudeAiPlugins` to `false` in Claude Code managed settings. Learn more about **[Claude Code admin setup](https://code.claude.com/docs/en/setup#advanced-setup)**.

**Note:** Marketplaces are how owners distribute plugins to their organization. Users can also share a plugin they built with specific colleagues or groups, or submit it to be published to your organization's library. Sharing and publishing are controlled by the settings in **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** in the “Policy” tab, and plugin submissions are reviewed there on the "Requests" tab. Learn more about **[using plugins in Claude](https://support.claude.com/en/articles/13837440-use-plugins-in-claude)** and **[letting users publish skills and plugins to your organization](https://support.claude.com/en/articles/13119606-provision-and-manage-skills-for-your-organization#h_1abc45a27c)**[.](https://support.claude.com/en/articles/13119606-provision-and-manage-skills-for-your-organization#h_1abc45a27c)

------------------------------------------------------------------------

## See what's in your organization

The “Inventory” tab lists every skill and plugin your organization governs, including the ones users created for themselves. Skills and plugins now live on the same page, so you don't need to switch between settings.

For each item, the table shows its source, version, capabilities, audience, and how many people used it in the last 30 days.

To find an item, search by name or use the filters:

- **Source:** Your organization, Organization library, or a specific marketplace

- **Audience:** Who can use the item

- **Type:** Skills or plugins

Skills your organization manages appear under **Organization library**. You manage plugin marketplaces in the **Marketplaces** tab.

To act on an item, click the menu button at the end of its row. You can view details and files, see version history, and change default and group access.

For skills and plugins that users created, the Inventory tab shows metadata and sharing status. It doesn't show the files themselves.

## Use Anthropic-built marketplaces

Anthropic provides built-in marketplaces of plugins for different functions, such as legal and finance. A Knowledge Work marketplace is added to your organization by default.

To add an Anthropic-built marketplace:

1. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Marketplaces” tab.

2. Click “Add,” then select “Add marketplace.”

3. Select "Browse Anthropic sources."

4. Click "Add" on each one you want to appear for everyone in your organization.

To remove the Knowledge Work marketplace if it isn't relevant to your teams:

1. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Marketplaces” tab.

2. Find **Knowledge Work**.

3. Click the menu button in the upper right corner.

4. Select "Remove."

Anthropic-built marketplaces show readable names, like "Knowledge Work," instead of the underlying repository slug.

------------------------------------------------------------------------

## Create your own marketplace

Beyond the Anthropic-built marketplaces, you can create your own. There are two ways to add plugins to a marketplace you create:

- **Manual upload:** Upload individual plugin ZIP files through the admin UI. Best for quick iteration, one-off tools, or teams that don't keep plugins in a Git repository.

- **Repository syncing:** Connect a private GitHub or GitLab repository and Claude syncs plugins from it. Best when multiple developers collaborate on plugins or you want version-controlled updates.

You can use both approaches in parallel. For example, run a GitHub-synced marketplace for your core plugins and a separate manual marketplace for ad-hoc tools.

### Set up a manual marketplace

1. Go to [](https://claude.ai/admin-settings/plugins) **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**.

2. Click “Add,” then select “Upload a plugin.”

3. If this is your first time setting up a marketplace, "Upload to a new marketplace" is auto-selected. You'll be able to choose "Add to an existing marketplace" when uploading plugins later.

4. Enter a name for your marketplace.

5. Either drag your files in, or click the upload prompt and select your file. The file must be a valid .zip under 200 MB.

6. Repeat for each plugin you want to add.

7. Click "Upload" to add your plugins to a new marketplace.

If you upload a plugin with the same name as an existing one, it overwrites the previous version automatically. You don't need to delete the old one first.

### Set up a GitHub-synced marketplace

**Prepare your repository**

You can connect a repo hosted on github.com or on your organization's GitHub Enterprise host. On github.com, your repository must be **private or internal**—public repos aren't allowed for organization marketplaces. To sync from GitLab, see **[Set up a GitLab-synced marketplace](#h_6be3dfbfbc)**.

A plugin source can be private in three cases:

- A github.com source that shares your marketplace repository's owner, which organization sync fetches through the Claude GitHub App.

- A source on your organization's GitHub Enterprise host with your organization's GitHub Enterprise App installed on that repository.

- A url or git-subdir source on the same GitLab host as your marketplace repository. On gitlab.com, the source must also be under the same top-level group or user namespace as the marketplace repository.

Every other source is fetched without credentials, so it must be a public repository on github.com, gitlab.com, or bitbucket.org. Sources on any other host are rejected.

If your plugin code lives in a private repository that doesn't meet the criteria above, copy those plugin folders into the marketplace repository and change each plugin's source to a relative path (a git subtree or a CI step that vendors the files works well).

For details on plugin structure and formatting, see the **[plugin manifest reference](https://code.claude.com/docs/en/plugins/manifest-reference)**.

Additional resources:

- **[Create a marketplace](https://code.claude.com/docs/en/plugins/create-marketplace)**

- **[Create a plugin](https://code.claude.com/docs/en/plugins/create)**

- **[Creating a new GitHub repository](https://docs.github.com/en/repositories/creating-and-managing-repositories/creating-a-new-repository)**

**Connect the repository**

1. Make sure both Cowork and Skills are enabled for your organization.

1. **[Enable Cowork](https://support.claude.com/en/articles/13455879-use-cowork-on-team-and-enterprise-plans#h_71cdc52dfc)**

2. **[Enable Skills](https://support.claude.com/en/articles/13119606-provision-and-manage-skills-for-your-organization#h_7673241237)**

2. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Marketplaces” tab.

3. Click “Add plugins,” then select “Sync from GitHub.”

4. Enter the repository in `owner/repo` format (for example, `acme-corp/claude-plugins`).

Your personal GitHub token is verified to confirm you have access, then Cowork uses its GitHub App installation token for sync operations.

**Can't see your repo?** Make sure the Claude GitHub App is installed in that repository.

**How syncing works**

An initial sync runs automatically when you connect a repository. In the GitHub creation flow, you can also turn off "Sync automatically," which is on by default. With it on, a sync runs when a pull request is merged into the branch the marketplace follows, or when someone pushes to that branch directly.

The person turning automatic sync on or off must have admin-level access to that repository on GitHub. This is checked through their personal GitHub connection, which is separate from the Claude GitHub App installation. Without admin access, the page shows "Cannot access repository. Ensure the repository exists and the Claude GitHub App is installed," even when the App is installed correctly and manual updates work.

Once enabled, automatic sync for a GitHub repository runs when a pull request that includes a plugin version bump is merged to the repository's default branch. Direct pushes to the default branch don't trigger a sync. You can always trigger a sync manually by clicking "Re-sync" on the marketplace.

**Important:** If a sync fails, members keep the last synced version of each plugin. Check the failure message, fix the repo, and re-sync.

### Set up a GitLab-synced marketplace

You can sync an organization marketplace from a repository on gitlab.com or on a self-managed GitLab instance. For supported plugin source types and which sources can be private, see **Prepare your repository** above.

GitLab support is in beta and applies to plugin marketplace sync only. Claude Code on the web doesn't support GitLab repositories yet.

**Before you start**

- **Role.** You need the Owner or Primary Owner role to add a GitLab configuration.

- **Project visibility.** On gitlab.com, the marketplace project must be private or internal. On a self-managed instance, any visibility works.

- **Access token.** Create a GitLab access token for Claude to read the marketplace project and any plugin source projects on the same host. Personal, group, and project access tokens all work. The token's user needs at least the Reporter role on those projects.

- Legacy (scoped) token: at least the `read_api` scope.

- Fine-grained personal access token: the Project: Read, Branch: Read, and Repository: Read permissions.

- Optional, for automatic webhook setup (step 3): the Maintainer role on the marketplace project plus the `api` scope (legacy) or the Webhook: Create and Delete permission (fine-grained). Without these, you add the webhook by hand once (step 4).

- **Self-managed instances only.**

- The instance's hostname must resolve to a public IP address, and the instance must accept HTTPS connections from **[Anthropic's outbound IP addresses](https://platform.claude.com/docs/en/api/ip-addresses#outbound-ip-addresses)**.

- If a private certificate authority issued its TLS certificate, have the CA certificate (PEM) ready.

- Webhook deliveries for automatic sync come from your instance to api.anthropic.com, so the instance must be able to make outbound HTTPS requests to that host.

**Step 1: Add a GitLab configuration**

To add a GitLab configuration:

1. Go to **[Organization settings \> Claude Code](https://claude.ai/admin-settings/claude-code)**.

2. Under **Self-hosted infrastructure**, find **GitLab** and click "Add configuration."

3. Enter a display name, the hostname (gitlab.com, or your instance's hostname such as gitlab.example.com), the access token, and, for a self-managed instance with a private CA, the CA certificate.

4. Click "Add configuration." Claude runs a connection test right away and shows the results in a **GitLab connection test** dialog.

5. (Optional) To check that the token can read your marketplace project, enter its path (for example, `platform/claude-plugins`) under **Project path** and click "Test sync access."

You can re-run the test at any time from the configuration's menu with **Test connection**. The hostname can't be changed later. If it changes, delete the configuration and add a new one.

**Step 2: Add the marketplace**

To add the marketplace:

1. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the "Marketplaces" tab.

2. Click "Add," then select "Sync from GitLab."

3. In **GitLab repository URL**, enter the project's HTTPS URL, for example [https://gitlab.example.com/platform/claude-plugins](https://gitlab.example.com/platform/claude-plugins). Projects in nested subgroups work.

4. Leave **Sync automatically** on if you want pushes to trigger a sync (see **Step 3**).

5. Choose the **Default access** for the plugins the marketplace contains.

6. Click "Create."

Claude reads the project's default branch and validates the marketplace file before creating the marketplace, which is named after the project path.

**Step 3: Automatic sync**

With **Sync automatically** on, every push to the project's default branch triggers a sync through a webhook on the GitLab project. If your token can manage webhooks (see **Before you start**), Claude creates the webhook for you and there's nothing else to do.

You can turn **Sync automatically** on or off later from the marketplace's menu. Turning it on sets up the webhook the same way.

**Step 4: If Claude can't create the webhook**

If the token can't manage webhooks, Claude shows a **Finish webhook setup in GitLab** message on screen with a **Webhook URL** and a token. The token is shown only once, so finish this step before closing the message.

1. In your GitLab project, open **Settings \> Webhooks** and add a new webhook with the URL from the message.

2. Paste the token into the field.

1. If the message shows a **Secret token**, paste it into GitLab's Secret token field.

2. If the message shows a **Signing token** (gitlab.com and self-managed versions 19.1 or later), select "Generate signing token" in GitLab and replace the generated value with the one from Claude.

3. Enable **Push events** and save.

4. (Optional) Use **Test \> Push events** on the new webhook in GitLab and confirm it reports HTTP 200.

If you lose the token, open **Configure webhook** from the marketplace's menu, click "Remove webhook," then "Enable webhook" to get a new one. The old token will stop working.

### Choose between manual upload and repository sync

[TABLE]

------------------------------------------------------------------------

## Control plugin distribution

Once your marketplace has plugins, you control how they're distributed using installation preferences. For each plugin, you can set one of four options:

[TABLE]

**Important:** Plugins set to "Installed by default" or "Required" also install in Claude Code for users who sign in with their Claude account. In Claude Code, the plugin's hooks, sub-agents, and MCP servers run on the user's computer, and a "Required" plugin can't be disabled there. Review a plugin's hooks before you set it to "Required." Learn more about **[synced plugins](https://code.claude.com/docs/en/plugins-reference#synced-plugins)** in the Claude Code docs.

### Set preferences

1. In **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**, click the “Inventory” tab.

2. Find the plugin and click the menu button at the end of its row.

3. Click “Default access,” then select an option under **Install**.

4. Changes take effect on each member's next session or plugin refresh.

### What members experience

Members browse available plugins on the **Discover** tab in **Customize \> Plugins**. Plugins you set to "Installed by default" or "Required" appear in their list automatically. Plugins you set to "Available to install" show up on **Discover** for members to add themselves.

Members can't edit organization-managed plugins, which prevents conflicting changes to shared tooling.

------------------------------------------------------------------------

## Customize plugin access by group

Group-level plugin access is available on Enterprise plans. You set it on the same **Organization settings \> Plugins & skills** page, which requires an Owner or a custom role that includes managing the organization's libraries.

### How group overrides work

Each plugin in your marketplace has an organization-wide installation preference (Installed by default, Available to install, Required, or Not available). By default, every group inherits that organization-wide setting.

When you set a group-level override for a plugin, it replaces the org-wide setting for members of that group. The resolution order is: group setting, then org-wide setting, then marketplace default.

### Set plugin access for a group

1. In [](https://claude.ai/admin-settings/plugins) **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**[,](https://claude.ai/admin-settings/skills) click the “Inventory” tab.

2. Find the plugin you want to customize and click the menu button at the end of its row.

3. Click “Group access…”.

4. Click “Add groups” next to **Install targeting**.

5. Select the group and choose one of the installation preferences listed above.

Both manually created groups and SCIM-provisioned groups from your identity provider appear in the group picker and work the same way.

### What happens when a member is in multiple groups

For example, if Group A sets a plugin to "Not available" and Group B sets it to "Installed by default," a member in both groups gets the plugin installed by default.

### What happens when a group is deleted

### Do group settings persist across marketplace re-syncs?

Yes. Group-level overrides persist when you re-sync a GitHub- or GitLab-connected marketplace. They're only removed if the plugin itself is deleted from the marketplace.

------------------------------------------------------------------------

## Update and remove plugins

### Manual marketplaces

To remove a plugin, delete it from your marketplace in **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** under the “Inventory” tab.

### GitHub-synced marketplaces

Push your changes to the connected repository, then go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Marketplaces” tab, find your marketplace, and click "Update" to trigger a sync. Each sync replaces all plugins with the current state of the repo. If an owner has enabled "Sync automatically" for the marketplace, a sync also runs whenever a pull request with a plugin version bump is merged to the repository's default branch.

For GitLab-synced marketplaces, see **[Set up a GitLab-synced marketplace](#h_6be3dfbfbc)**.

To remove a plugin, delete it from the repository and trigger a sync.

------------------------------------------------------------------------

## Limits

[TABLE]

------------------------------------------------------------------------

## Naming rules

Plugin names must use **lowercase words separated by hyphens** (for example, `deployment-tools`, not `Deployment Tools`). The following marketplace names are reserved and can't be used:

- `claude-code-marketplace`

- `claude-code-plugins`

- `claude-plugins-official`

- `anthropic-marketplace`

- `anthropic-plugins`

- `agent-skills`

- `life-sciences`

Names that impersonate official Anthropic marketplaces are also blocked.

------------------------------------------------------------------------

## Troubleshooting

### Upload rejected

Common causes: the file exceeds 200 MB, it isn't a valid ZIP file, or the marketplace has reached the 1000-plugin limit. Check the file size and format, and remove unused plugins if you're at capacity. If skill and plugin scanning is on, a plugin can also be rejected because it was flagged for malicious content. Review the reason, fix the issue, and upload the plugin again.

### Plugin not appearing for members

### Updated plugin not reflecting for members

Changes take effect on each member's next session or plugin refresh. In Claude Code, users get the change the next time they start Claude Code. If the update still isn't showing, confirm the upload succeeded by checking the plugin version in your marketplace.

### GitHub sync fails with a content error

One or more plugins in your repo is likely formatted incorrectly. Fix the formatting issue, push the update to GitHub, and trigger the sync again. For plugin structure requirements, see the **[plugin manifest reference](https://code.claude.com/docs/en/plugins/manifest-reference)**.

### Sync fails with "External plugin source type is not supported. Supported types: git-subdir, github, url" or plugins are skipped with "Repository not found on gitlab.com. Check the URL and make sure the repository is public."

One or more plugin entries in your `marketplace.json` use a `source` that points outside the connected repository (a `github`, `url`, or `git-subdir` source), and organization sync can't fetch it. A private source only works in three cases: a github.com repository shares your marketplace repository's owner, a repository on your organization's GitHub Enterprise host with your GitHub Enterprise App installed on it, or a url or git-subdir source on the same GitLab host as your marketplace repository (on gitlab.com, also under the same top-level group or user namespace).

For any other private source, move the plugin folders into the marketplace repository and change each entry's `source` to a relative path (for example, `"./plugins/my-plugin"`), then push and re-sync. Alternatively, upload the affected plugins individually via **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** Click “Add,” “Upload a plugin,” then select "Add to an existing marketplace." Plugins uploaded through a member's own Customize menu are installed only for that member and aren't distributed to your organization, although members can share plugins with specific colleagues or groups if sharing is turned on.

### Can't see a GitHub repo when connecting

Make sure the Claude GitHub App is installed in that repository. Your personal GitHub token is checked first to confirm access, but the sync itself uses the GitHub App installation token.

### "Cannot access repository" when turning on "Sync automatically"

If manual updates work but turning on "Sync automatically" shows "Cannot access repository. Ensure the repository exists and the Claude GitHub App is installed," there are two likely causes:

### "No GitLab instance is configured for your organization yet. Add one under Claude Code settings, then come back here" when adding a marketplace

If you see this message when adding a marketplace in GitLab, make sure you’ve added the GitLab configuration first. See **Step 1: Add a GitLab configuration** under **[Set up a GitLab-synced marketplace](#h_6be3dfbfbc)**.

### "This URL isn’t on a configured GitLab instance"

Make sure the URL's hostname matches a configured GitLab hostname exactly.

### Adding a gitlab.com project fails with "Failed to create marketplace. Try again."

Check that the project's visibility is private or internal. Public gitlab.com projects can't be used as organization marketplaces.

### The connection test fails at Reach host (for example "The hostname couldn’t be resolved," "The connection was refused," or "GitLab didn’t respond in time"), or adding a marketplace on a self-managed instance fails with a generic error

Confirm the instance is reachable from the public internet over HTTPS and that Anthropic's outbound IP addresses are allowed through your firewall. "A secure connection couldn’t be established" usually means the CA certificate is missing or wrong.

### Errors that start with "Authentication to … failed" or "Access denied"

The token is expired, revoked, or missing a scope or role listed in the **Before you start** section under **[Set up a GitLab-synced marketplace](#h_6be3dfbfbc)**. Create a new token and paste it into the configuration (**Edit \> Access token**).

---
## manage-cowork-plugins-for-your-organization.md

4. Manage Cowork plugins for your organization

# Manage Cowork plugins for your organization

Plugin marketplaces allow Team and Enterprise plan owners to distribute curated plugins to everyone in their organization through Cowork. You create a marketplace, add plugins to it, and control exactly which plugins your team members can see and use.

Owners and Primary Owners of Team and Enterprise plans can manage organization plugins on Claude Desktop.

**Requirements:** Cowork and Skills must both be enabled for your organization before you can use plugin marketplaces.

There are two ways to add plugins to a marketplace:

- **Manual upload**—Upload individual plugin ZIP files through the admin UI. Best for quick iteration, one-off tools, or teams that don't use GitHub for plugin development.

- **GitHub syncing**—Connect a private GitHub repository and Cowork automatically syncs plugins from it. Best when multiple developers collaborate on plugins or you want version-controlled updates.

You can use both approaches in parallel—for example, a GitHub-synced marketplace for your core plugins and a separate manual marketplace for ad-hoc tools.

------------------------------------------------------------------------

## Set up a manual marketplace

1. Open Claude Desktop and go to **Organization settings \> Plugins**.

2. Click “Add plugins” and select “Upload a file” as the source.

3. If this is your first time setting up a marketplace, “Upload to a new marketplace” will be auto-selected.

1. You’ll be able to choose “Add to an existing marketplace” when uploading plugins moving forward.

4. Enter a name for your marketplace.

5. Either drag your files in, or click the upload prompt and select your file.

1. **Note:** The file must be a valid .zip under 50 MB.

6. Repeat for each plugin you want to add.

7. Click “Upload” to add your plugins to a new marketplace.

If you upload a plugin with the same name as an existing one, it overwrites the previous version automatically—no need to delete the old one first.

------------------------------------------------------------------------

## Set up a GitHub-synced marketplace

### Prepare your repository

Your repository must be **private or internal**—public repos aren't allowed for organization marketplaces. Repos hosted on custom GitHub Enterprise Server instances aren't supported. Your repo must be hosted on github.com.

For details on plugin structure and formatting, see the **[plugin reference documentation](../08-Plugins-Skills/plugins-reference-claude-code-docs.md)**.

Additional resources:

- **[Create and distribute a plugin marketplace](../08-Plugins-Skills/create-and-distribute-a-plugin-marketplace-claude-code-docs-34a8f3c5e8.md)**

- **[Create plugins](../08-Plugins-Skills/create-plugins-claude-code-docs.md)**

- **[Creating a new GitHub repository](https://docs.github.com/en/repositories/creating-and-managing-repositories/creating-a-new-repository)**

### Connect the repository

1. Make sure both Cowork and Skills are enabled for your organization.

2. Go to **Organization settings \> Plugins**.

3. Click "Add plugin" and select "GitHub" as the source.

4. Enter the repository in `owner/repo` format (for example, `acme-corp/claude-plugins`).

Your personal GitHub token is verified to confirm you have access, then Cowork uses its GitHub App installation token for sync operations.

**Can't see your repo?** Make sure the Claude GitHub App is installed in that repository.

### How syncing works

An initial sync runs automatically when you connect a repository. After that, you can trigger syncs manually by clicking “Check for updates” on your marketplace in **Organization settings \> Plugins**.

------------------------------------------------------------------------

## Control plugin distribution

Once your marketplace has plugins, you control how they're distributed using installation preferences. For each plugin, you can set one of four options:

[TABLE]

### Set preferences

1. In **Organization settings \> Plugins**, navigate to your marketplace.

2. Select the installation preference for each plugin.

3. Changes take effect on each member's next session or plugin refresh.

### What members experience

Members browse available plugins through the **Browse plugins** modal in Cowork. Auto-installed plugins appear in their installed list automatically. Available plugins show up in the catalog for self-service installation.

Members can't edit organization-managed plugins, which prevents conflicting changes to shared tooling.

------------------------------------------------------------------------

## Update and remove plugins

### Manual marketplaces

To update a plugin, upload a new ZIP file with the same plugin name. The new version overwrites the existing one automatically. Plugin names are the unique identifier — `legal` will always replace `legal`.

To remove a plugin, delete it from your marketplace in **Organization settings \> Plugins**.

### GitHub-synced marketplaces

Push your changes to the connected repository, then go to **Organization settings \> Plugins**, find your marketplace, and click "Check for updates" to trigger a sync. Each sync replaces all plugins with the current state of the repo.

To remove a plugin, delete it from the repository and trigger a sync.

------------------------------------------------------------------------

## Limits

[TABLE]

------------------------------------------------------------------------

## Naming rules

Plugin names must use **lowercase words separated by hyphens** (for example, `deployment-tools`, not `Deployment Tools`). The following marketplace names are reserved and can't be used:

- `claude-code-marketplace`

- `claude-code-plugins`

- `claude-plugins-official`

- `anthropic-marketplace`

- `anthropic-plugins`

- `agent-skills`

- `life-sciences`

Names that impersonate official Anthropic marketplaces are also blocked.

------------------------------------------------------------------------

## Choose between manual upload and GitHub sync

[TABLE]

------------------------------------------------------------------------

## Troubleshooting

### Upload rejected

Common causes: the file exceeds 50 MB, it isn't a valid ZIP file, or the marketplace has reached the 100-plugin limit. Check the file size and format, and remove unused plugins if you're at capacity.

### Plugin not appearing for members

### Updated plugin not reflecting for members

Changes take effect on each member's next session or plugin refresh. If the update still isn't showing, confirm the upload succeeded by checking the plugin version in your marketplace.

### GitHub sync fails with a content error

One or more plugins in your repo is likely formatted incorrectly. Fix the formatting issue, push the update to GitHub, and trigger the sync again. For plugin structure requirements, see the **[plugin reference documentation](../08-Plugins-Skills/plugins-reference-claude-code-docs.md)**.

### Plugins disappeared after a failed sync

### Can't see a GitHub repo when connecting

Make sure the Cowork GitHub App is installed in that repository. Your personal GitHub token is checked first to confirm access, but the sync itself uses the GitHub App installation token.

---
## manage-model-access-for-your-organization-13c8d61731.md

# Manage model access for your organization

Copy for LLM

This guide explains how to control which Claude models members of your organization can use, and how to cap the effort level members can select on each model. Model access and effort limits can be set for your whole organization or for specific custom roles.

Model access settings are available for Enterprise plan organizations. Primary Owners, Owners, and members whose custom role grants the Identity & Access permission can manage them in **[Organization settings \> Models](https://claude.ai/admin-settings/models)**.

To set the model new conversations start on, see **[Set a default model for your organization](https://support.claude.com/en/articles/15330088)**.

------------------------------------------------------------------------

## How model access works

Model access and effort limits are determined at two levels:

- **Organization level:** each model is enabled or disabled for everyone in your organization. Disabling a model here removes it for every member, including Owners and Admins. You can also set a maximum effort level for each enabled model, which applies to every member.

- **Custom role level:** for members on custom roles, each role grants access to a subset of the models enabled at the organization level. A role can also cap the maximum effort level members can select on each model, at or below the organization's cap for that model.

The organization setting is the ceiling. A role can't grant access to a model that's disabled for the organization, and a role can't allow an effort level higher than the organization's cap. When the feature first becomes available, every model is enabled and set to its highest effort level at both levels, so nothing changes for your members until you adjust these settings.

**Note:** Haiku models are always available to every member and can’t be disabled. This guarantees members always have at least one model to fall back to.

## Who each level affects

- Disabling a model or capping its effort level at the organization level affects every member, including Primary Owners, Owners, Admins, and Users.

- Role-level model access and effort limits affect only members whose role is set to "Custom." Members with the User, Admin, or Owner roles can use every model enabled at the organization level, up to the organization's effort cap for that model.

------------------------------------------------------------------------

## Enable or disable a model for your organization

1. Navigate to **[Organization settings \> Models](https://claude.ai/admin-settings/models)**.

2. Under **Model access**, find the model you want to change.

3. To enable a model, switch the toggle next to it on, then click the role dropdown to select the roles that can access it.

4. To disable a model, click the role dropdown and deselect the roles before switching the toggle off.

5. Click "Save."

If any custom role uses the model you’re disabling as its default, you’ll be prompted to change that role’s default before the change can be saved.

------------------------------------------------------------------------

## Limit the maximum effort level for your organization

Effort limits determine how much computation members can apply per response on each model. Higher effort levels produce more thorough responses but consume more usage. An organization-level effort cap applies to every member and is the highest level any custom role can allow.

1. Navigate to **[Organization settings \> Model](https://claude.ai/admin-settings/models)**[s](https://claude.ai/admin-settings/models).

2. Under **Model access**, find the model you want to change.

3. Click the effort level dropdown to select the maximum level.

4. Click "Save."

If any custom role has an effort cap higher than the new organization cap for that model, the role's cap is lowered to match. Members see only effort levels at or below the organization cap in their model menu. Available effort levels differ depending on the model, and some models don't support effort level settings at all. For an explanation of each level, see **[Change the model, effort, and thinking settings](https://support.claude.com/en/articles/8664678)**.

------------------------------------------------------------------------

## Set model access for a custom role

1. Navigate to **[Organization settings \> Roles](https://claude.ai/admin-settings/roles)**.

2. Click the role you want to edit, or click “Add role” to create one.

3. Select the "Models" tab.

4. Under **Model access**, switch each model on or off. Models disabled at the organization level appear but can’t be enabled until you turn them on for the organization.

5. Click “Save.”

Only models the role grants access to can be selected as that role’s default model.

------------------------------------------------------------------------

## Limit the maximum effort level for a custom role

Effort limits determine how much computation members on a role can apply per response on each model. Higher effort levels produce more thorough responses but consume more usage. A role's effort cap can't be higher than the organization's cap for that model.

1. Navigate to **[Organization settings \> Roles](https://claude.ai/admin-settings/roles)**.

2. Click the role you want to edit.

3. Select the "Models" tab.

4. Next to a model, click the gear icon and choose a level.

5. Click "Save" to save your changes.

Members on the role see only effort levels at or below the cap in their model menu. Note that available effort levels differ depending on the model, and some models don’t support effort level settings at all. For an explanation of each level, see **[Change the model, effort, and thinking settings](https://support.claude.com/en/articles/8664678)**.

------------------------------------------------------------------------

## How access combines across multiple roles

If a member belongs to multiple groups with different custom roles, model settings combine like other role permissions:

- **Model access is additive.** The member can use every model granted by any of their roles, as long as it’s enabled at the organization level.

- **Effort limits take the highest cap.** For each model, the member gets the highest maximum effort level any of their roles allows, never exceeding the organization's cap for that model.

For how default models are chosen across multiple roles, see **[Set a default model for your organization](https://support.claude.com/en/articles/15330088)**.

For details on creating roles and assigning them to groups, see **[Manage custom roles on Enterprise plans](https://support.claude.com/en/articles/13930452)**.

------------------------------------------------------------------------

## What users see

In every covered product, the model picker shows only the models the member has access to. Effort levels above the organization's cap, or above a role's cap, don't appear in the effort menu.

Model availability also depends on the product. Each product supports a different set of models, so an enabled model appears only in the products that support it.

If you disable a model a member is using in an open conversation or session, that conversation falls back to the member's default model the next time they open it. If the member sends a message while you're making the change, they'll see an error that the model isn't available and be prompted to switch. If you lower a model's effort cap while a member has a higher level selected, their next message on that model uses the new maximum.

------------------------------------------------------------------------

## Where model access settings apply

Model access settings are enforced across these products:

[TABLE]

**Note:** Members on Claude Code CLI versions earlier than 2.1.199 still see disabled models and effort levels in the picker, but requests using them are rejected.

## Model access and Claude Code managed settings

If your organization also configures Claude Code through `managed-settings.json`, the `availableModels` setting and model access work together. In Claude Code CLI and IDE, members see only models that appear in `availableModels` *and* are enabled by their model access settings—a model removed by either one is unavailable.

Managed settings for models apply only to Claude Code CLI and IDE, not to Claude Code on web or desktop. For consistent behavior across all Claude Code surfaces, we recommend using model access settings alone. For more on managed settings, see **[Claude Code settings](https://code.claude.com/docs/en/settings#settings-files)**.

---
## manage-plugins-for-your-organization-af9f814bdd.md

# Manage plugins for your organization

Copy for LLM

Plugin marketplaces let Team and Enterprise plan owners distribute curated plugins to everyone in their organization. You create a marketplace, add plugins to it, and control exactly which plugins your team members can see and use. Plugins you distribute appear in chat (on the web and the Chat tab in Claude Desktop), in Claude Cowork, and in Claude Code sessions where members sign in with the same Claude account.

**Requirements:** Cowork and Skills must both be enabled for your organization before you can use plugin marketplaces.

For a reference version of these controls, see **[Manage plugins for your organization](https://code.claude.com/docs/en/plugins/org)** in the Claude Code docs.

------------------------------------------------------------------------

## See what's in your organization

For each item, the table shows its source, version, capabilities, audience, and how many people used it in the last 30 days.

To find an item, search by name or use the filters:

- **Source:** Your organization, Organization library, or a specific marketplace

- **Audience:** Who can use the item

- **Type:** Skills or plugins

Skills your organization manages appear under **Organization library**. You manage plugin marketplaces in the **Marketplaces** tab.

To act on an item, click the menu button at the end of its row. You can view details and files, see version history, and change default and group access.

For skills and plugins that users created, the Inventory tab shows metadata and sharing status. It doesn't show the files themselves.

## Use Anthropic-built marketplaces

Anthropic provides built-in marketplaces of plugins for different functions, such as legal and finance. A Knowledge Work marketplace is added to your organization by default.

To add an Anthropic-built marketplace:

1. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Marketplaces” tab.

2. Click “Add,” then select “Add marketplace.”

3. Select "Browse Anthropic sources."

4. Click "Add" on each one you want to appear for everyone in your organization.

To remove the Knowledge Work marketplace if it isn't relevant to your teams:

1. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Marketplaces” tab.

2. Find **Knowledge Work**.

3. Click the menu button in the upper right corner.

4. Select "Remove."

Anthropic-built marketplaces show readable names, like "Knowledge Work," instead of the underlying repository slug.

------------------------------------------------------------------------

## Create your own marketplace

Beyond the Anthropic-built marketplaces, you can create your own. There are two ways to add plugins to a marketplace you create:

- **Manual upload:** Upload individual plugin ZIP files through the admin UI. Best for quick iteration, one-off tools, or teams that don't keep plugins in a Git repository.

- **Repository syncing:** Connect a private GitHub or GitLab repository and Claude syncs plugins from it. Best when multiple developers collaborate on plugins or you want version-controlled updates.

You can use both approaches in parallel. For example, run a GitHub-synced marketplace for your core plugins and a separate manual marketplace for ad-hoc tools.

### Set up a manual marketplace

1. Go to [](https://claude.ai/admin-settings/plugins) **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**.

2. Click “Add,” then select “Upload a plugin.”

3. If this is your first time setting up a marketplace, "Upload to a new marketplace" is auto-selected. You'll be able to choose "Add to an existing marketplace" when uploading plugins later.

4. Enter a name for your marketplace.

5. Either drag your files in, or click the upload prompt and select your file. The file must be a valid .zip under 200 MB.

6. Repeat for each plugin you want to add.

7. Click "Upload" to add your plugins to a new marketplace.

If you upload a plugin with the same name as an existing one, it overwrites the previous version automatically. You don't need to delete the old one first.

### Set up a GitHub-synced marketplace

**Prepare your repository**

A plugin source can be private in three cases:

- A github.com source that shares your marketplace repository's owner, which organization sync fetches through the Claude GitHub App.

- A source on your organization's GitHub Enterprise host with your organization's GitHub Enterprise App installed on that repository.

Every other source is fetched without credentials, so it must be a public repository on github.com, gitlab.com, or bitbucket.org. Sources on any other host are rejected.

For details on plugin structure and formatting, see the **[plugin manifest reference](https://code.claude.com/docs/en/plugins/manifest-reference)**.

Additional resources:

- **[Create a marketplace](https://code.claude.com/docs/en/plugins/create-marketplace)**

- **[Create a plugin](https://code.claude.com/docs/en/plugins/create)**

- **[Creating a new GitHub repository](https://docs.github.com/en/repositories/creating-and-managing-repositories/creating-a-new-repository)**

**Connect the repository**

1. Make sure both Cowork and Skills are enabled for your organization.

1. **[Enable Cowork](https://support.claude.com/en/articles/13455879-use-cowork-on-team-and-enterprise-plans#h_71cdc52dfc)**

2. **[Enable Skills](https://support.claude.com/en/articles/13119606-provision-and-manage-skills-for-your-organization#h_7673241237)**

2. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Marketplaces” tab.

3. Click “Add plugins,” then select “Sync from GitHub.”

4. Enter the repository in `owner/repo` format (for example, `acme-corp/claude-plugins`).

Your personal GitHub token is verified to confirm you have access, then Cowork uses its GitHub App installation token for sync operations.

**Can't see your repo?** Make sure the Claude GitHub App is installed in that repository.

**How syncing works**

**Important:** If a sync fails, members keep the last synced version of each plugin. Check the failure message, fix the repo, and re-sync.

### Set up a GitLab-synced marketplace

GitLab support is in beta and applies to plugin marketplace sync only. Claude Code on the web doesn't support GitLab repositories yet.

**Before you start**

- **Role.** You need the Owner or Primary Owner role to add a GitLab configuration.

- **Project visibility.** On gitlab.com, the marketplace project must be private or internal. On a self-managed instance, any visibility works.

- Legacy (scoped) token: at least the `read_api` scope.

- Fine-grained personal access token: the Project: Read, Branch: Read, and Repository: Read permissions.

- **Self-managed instances only.**

- If a private certificate authority issued its TLS certificate, have the CA certificate (PEM) ready.

- Webhook deliveries for automatic sync come from your instance to api.anthropic.com, so the instance must be able to make outbound HTTPS requests to that host.

**Step 1: Add a GitLab configuration**

To add a GitLab configuration:

1. Go to **[Organization settings \> Claude Code](https://claude.ai/admin-settings/claude-code)**.

2. Under **Self-hosted infrastructure**, find **GitLab** and click "Add configuration."

3. Enter a display name, the hostname (gitlab.com, or your instance's hostname such as gitlab.example.com), the access token, and, for a self-managed instance with a private CA, the CA certificate.

4. Click "Add configuration." Claude runs a connection test right away and shows the results in a **GitLab connection test** dialog.

5. (Optional) To check that the token can read your marketplace project, enter its path (for example, `platform/claude-plugins`) under **Project path** and click "Test sync access."

You can re-run the test at any time from the configuration's menu with **Test connection**. The hostname can't be changed later. If it changes, delete the configuration and add a new one.

**Step 2: Add the marketplace**

To add the marketplace:

1. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the "Marketplaces" tab.

2. Click "Add," then select "Sync from GitLab."

4. Leave **Sync automatically** on if you want pushes to trigger a sync (see **Step 3**).

5. Choose the **Default access** for the plugins the marketplace contains.

6. Click "Create."

Claude reads the project's default branch and validates the marketplace file before creating the marketplace, which is named after the project path.

**Step 3: Automatic sync**

You can turn **Sync automatically** on or off later from the marketplace's menu. Turning it on sets up the webhook the same way.

**Step 4: If Claude can't create the webhook**

1. In your GitLab project, open **Settings \> Webhooks** and add a new webhook with the URL from the message.

2. Paste the token into the field.

1. If the message shows a **Secret token**, paste it into GitLab's Secret token field.

2. If the message shows a **Signing token** (gitlab.com and self-managed versions 19.1 or later), select "Generate signing token" in GitLab and replace the generated value with the one from Claude.

3. Enable **Push events** and save.

4. (Optional) Use **Test \> Push events** on the new webhook in GitLab and confirm it reports HTTP 200.

If you lose the token, open **Configure webhook** from the marketplace's menu, click "Remove webhook," then "Enable webhook" to get a new one. The old token will stop working.

### Choose between manual upload and repository sync

[TABLE]

------------------------------------------------------------------------

## Control plugin distribution

Once your marketplace has plugins, you control how they're distributed using installation preferences. For each plugin, you can set one of four options:

[TABLE]

### Set preferences

1. In **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**, click the “Inventory” tab.

2. Find the plugin and click the menu button at the end of its row.

3. Click “Default access,” then select an option under **Install**.

4. Changes take effect on each member's next session or plugin refresh.

### What members experience

Members can't edit organization-managed plugins, which prevents conflicting changes to shared tooling.

------------------------------------------------------------------------

## Customize plugin access by group

### How group overrides work

When you set a group-level override for a plugin, it replaces the org-wide setting for members of that group. The resolution order is: group setting, then org-wide setting, then marketplace default.

### Set plugin access for a group

2. Find the plugin you want to customize and click the menu button at the end of its row.

3. Click “Group access…”.

4. Click “Add groups” next to **Install targeting**.

5. Select the group and choose one of the installation preferences listed above.

Both manually created groups and SCIM-provisioned groups from your identity provider appear in the group picker and work the same way.

### What happens when a member is in multiple groups

For example, if Group A sets a plugin to "Not available" and Group B sets it to "Installed by default," a member in both groups gets the plugin installed by default.

### What happens when a group is deleted

### Do group settings persist across marketplace re-syncs?

Yes. Group-level overrides persist when you re-sync a GitHub- or GitLab-connected marketplace. They're only removed if the plugin itself is deleted from the marketplace.

------------------------------------------------------------------------

## Update and remove plugins

### Manual marketplaces

To remove a plugin, delete it from your marketplace in **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** under the “Inventory” tab.

### GitHub-synced marketplaces

For GitLab-synced marketplaces, see **[Set up a GitLab-synced marketplace](#h_6be3dfbfbc)**.

To remove a plugin, delete it from the repository and trigger a sync.

------------------------------------------------------------------------

## Limits

[TABLE]

------------------------------------------------------------------------

## Naming rules

Plugin names must use **lowercase words separated by hyphens** (for example, `deployment-tools`, not `Deployment Tools`). The following marketplace names are reserved and can't be used:

- `claude-code-marketplace`

- `claude-code-plugins`

- `claude-plugins-official`

- `anthropic-marketplace`

- `anthropic-plugins`

- `agent-skills`

- `life-sciences`

Names that impersonate official Anthropic marketplaces are also blocked.

------------------------------------------------------------------------

## Troubleshooting

### Upload rejected

### Plugin not appearing for members

### Updated plugin not reflecting for members

### GitHub sync fails with a content error

One or more plugins in your repo is likely formatted incorrectly. Fix the formatting issue, push the update to GitHub, and trigger the sync again. For plugin structure requirements, see the **[plugin manifest reference](https://code.claude.com/docs/en/plugins/manifest-reference)**.

### Can't see a GitHub repo when connecting

Make sure the Claude GitHub App is installed in that repository. Your personal GitHub token is checked first to confirm access, but the sync itself uses the GitHub App installation token.

### "Cannot access repository" when turning on "Sync automatically"

If manual updates work but turning on "Sync automatically" shows "Cannot access repository. Ensure the repository exists and the Claude GitHub App is installed," there are two likely causes:

### "No GitLab instance is configured for your organization yet. Add one under Claude Code settings, then come back here" when adding a marketplace

### "This URL isn’t on a configured GitLab instance"

Make sure the URL's hostname matches a configured GitLab hostname exactly.

### Adding a gitlab.com project fails with "Failed to create marketplace. Try again."

Check that the project's visibility is private or internal. Public gitlab.com projects can't be used as organization marketplaces.

### Errors that start with "Authentication to … failed" or "Access denied"

---
## manage-project-visibility-and-sharing-0a8cd4d8a3.md

# Manage project visibility and sharing

Copy for LLM

Project visibility and sharing features are available to users on Team or Enterprise plans. This article assumes that public projects and project sharing are enabled for your organization. Admins can turn either off. For more information, see **[Control project sharing for your organization](https://support.claude.com/en/articles/9927533)**.

Sharing a project with a group requires an Enterprise plan and is currently in beta.

## Understand the two project visibility options

When creating a project on a Team or Enterprise plan, you can choose between two initial visibility options:

- **Public:** Everyone in your organization can view and use the project.

- **Private:** Only invited members can view and use the project.

## What are public projects?

If you choose to share a project with the rest of your organization upon creation, this means that any other member of your team can find it by browsing or searching the **[Organization tab in the Projects section](https://claude.ai/projects?category=team)**, and use it to start a chat. Even if a project is public, your chats within that project will be private and inaccessible to other members of your organization unless you manually share them.

### Can I change a project from public to private?

Yes, you can switch the visibility of a project you created as public to private at any time by opening the project and clicking the “Share” button to the right of the project name.

Click “Everyone at \[your organization\]” under **General access** and select “Only people invited” to change the project from public to private.

## What are private projects?

Choosing “Only people invited” keeps your project private so that you are the only user with access by default. All content within the project (chats, project knowledge, and artifacts) remains inaccessible to other members unless they are explicitly invited.

### Can I change a project from private to public?

Yes, you can switch the visibility of a project you created as private to public at any time by opening the project and clicking the “Share” button to the right of the project name.

Click “Only people invited” under General access and select “Everyone at \[your organization\]” to change the project from private to public.

## Add and remove access to private projects

### To add members to a private project

1. Click the "Share" button to the right of the project name.

2. In the window that appears, add people from your organization by name or email address. On Enterprise plans, you can also add a group if its visibility settings allow project sharing. Groups appear in the results with a group icon and the label **Group**.

3. Select the appropriate permission level:

- **Can view:** Members can see project contents, knowledge, and instructions, and chat within the project, but cannot edit it.

- **Can edit:** Members can modify project instructions and knowledge, update member settings, and actively contribute to the project.

This will share the project and knowledge base with the member, but your chats will remain private unless you choose to share them.

### To remove members from a private project

1. A project member with "Can edit" permissions can open the project's "Share" menu.

2. Click on the role to the right of the member's name.

3. Select "Remove access."

4. After being removed, they will no longer be able to access this project or its content.

### Add members in bulk

You can add multiple users at once by copying and pasting a list of email addresses into the **Add people or groups** field after clicking “Share."

### Share a project with a group

Sharing a project with a group is currently in beta for Enterprise plans.

On Enterprise plans, you can share a project with a group instead of adding each member individually. Access follows group membership, so members who join the group later get access automatically, and members who leave lose it.

Before you can share with a group, an admin needs to turn on **Share resources with this group** in the group's visibility settings. See **[Manage groups and group spend limits on Enterprise plans](https://support.claude.com/en/articles/13799932-manage-groups-and-group-spend-limits-on-enterprise-plans)**.

1. Click the "Share" button to the right of the project name.

2. Type the group name in the **Add people or groups** field.

3. Select the group from the results.

4. Select "Can view" or "Can edit."

5. Click "Share."

A few things to know:

- A newly visible group can take a few minutes to appear in the picker.

- Access changes can take up to five minutes to apply.

- Group members only get access if they're also members of the organization the project belongs to. A group can span child organizations, but a shared project stays scoped to its own organization.

- Once a project is shared with a group, every member of that group can open it, even if the group isn't visible to their organization. Visibility controls who can find a group in the picker, not who can reach a project that's already shared.

- If a member leaves a group, they lose access to projects shared through it unless they have access another way.

- Deleting a group removes its access to any projects shared with it.

### Email notifications

When someone shares a project with you, you'll receive an email notification letting you know about the new shared project and your permission level.

### “Shared with you” tab

Projects that have been shared with you will appear in a "Shared with you" tab on your **[Projects section](https://claude.ai/projects?category=shared_with_me)**, making it easy to find and access collaborative work.

### What happens when archiving a shared project?

Archiving a project doesn't reset its sharing permissions or remove members. All members, permission levels, and project knowledge are preserved, and everything is restored exactly as it was when you unarchive the project. To revoke someone's access, remove them in the project's sharing settings before or after archiving.

## If you can't share projects

Owners can turn off project sharing for your organization. On Enterprise plans, they can also turn it off for specific roles. When sharing is off for you:

- You can still open a project's “Share” menu, but you'll see "Project sharing is turned off by your administrator." You can't add new users or groups.

- You can still change or remove existing access.

- Projects that are already shared stay shared, and users who already have access keep it.

- Public projects become private, so people who only had access because a project was public lose it. They stay private if your admin turns sharing back on.

Contact your admin if you need to share a project.

## Share and unshare chats

Chats within a project are not shared by default.

### To share a chat

1. Click the "Share" button in the upper right corner of your chat.

2. Click the "Share" button in the pop-up to create a shareable link.

The chat snapshot includes all messages that were sent prior to sharing the chat, including any artifacts. All messages sent after sharing a chat will remain private by default. However, you can choose to update the shared snapshot to include new messages.

### To update a shared chat

1. Open the "Share" menu.

2. Click “Update" next to “New messages since last shared” to include any new messages in the shared snapshot.

### To unshare a chat

1. Navigate to the "Share" menu.

2. Click the visibility dropdown.

3. Change the chat from "Shared" to "Private" to disable the direct link.

---
## manage-project-visibility-and-sharing.md

4. Manage project visibility and sharing

# Manage project visibility and sharing

Project visibility and sharing features are available to users on Team or Enterprise plans. This article assumes that public projects are enabled for your organization. For more information, see **[How can I disable public projects?](https://support.claude.com/en/articles/9927533-how-can-i-disable-public-projects)**

## Understand the two project visibility options

When creating a project on a Team or Enterprise plan, you can choose between two initial visibility options:

- **Public:** Everyone in your organization can view and use the project.

- **Private:** Only invited members can view and use the project.

## What are public projects?

If you choose to share a project with the rest of your organization upon creation, this means that any other member of your team can find it by browsing or searching the [Team tab in the Projects section](https://claude.ai/projects?category=team), and use it to start a chat. Even if a project is public, your chats within that project will be private and inaccessible to other members of your organization unless you manually share them.

### Can I change a project from public to private?

Yes, you can switch the visibility of a project you created as public to private at any time by opening the project and clicking the “Share” button to the right of the project name:

Click “Everyone at \[your organization\]” under **General access** and select “Only people invited” to change the project from public to private:

## What are private projects?

### Can I change a project from private to public?

Yes, you can switch the visibility of a project you created as private to public at any time by opening the project and clicking the “Share” button to the right of the project name:

Click “Only people invited” under General access and select “Everyone at \[your organization\]” to change the project from private to public:

## Add and remove member access to private projects

### To add members to a private project

1. Click the "Share" button to the right of the project name.

2. In the window that appears, add individual members from your organization using their names or email addresses.

3. Select the appropriate permission level:

- **Can view:** Members can see project contents, knowledge, and instructions, and chat within the project, but cannot edit it.

- **Can edit:** Members can modify project instructions and knowledge, update member settings, and actively contribute to the project.

This will share the project and knowledge base with the member, but your chats will remain private unless you choose to share them.

### To remove members from a private project

1. A project member with "Can edit" permissions can open the project's "Share" menu.

2. Click on the role to the right of the member's name.

3. Select "Remove access."

4. After being removed, they will no longer be able to access this project or its content.

### Add members in bulk

You can add multiple users at once by copying and pasting a list of email addresses into the **Invite by email** field after clicking “Share”:

### Email notifications

When someone shares a project with you, you'll receive an email notification letting you know about the new shared project and your permission level.\

### “Shared with me” tab

Projects that have been shared with you will appear in a "Shared with you" tab on your [Projects section](https://claude.ai/projects?category=shared_with_me), making it easy to find and access collaborative work.

### What happens when archiving a shared project?

When a project is archived, all sharing permissions are reset to private and previous sharing context is wiped for security.

## Share and unshare chats

Chats within a project are not shared by default.

### To share a chat

1. Click the "Share" button in the upper right corner of your chat.

2. Click the "Share" button in the pop-up to create a shareable link.

### To update a shared chat

1. Open the "Share" menu.

2. Click “Update" next to “New messages since last shared” to include any new messages in the shared snapshot.

### To unshare a chat

1. Navigate to the "Share" menu.

2. Click the visibility dropdown.

3. Change the chat from "Shared" to "Private" to disable the direct link.

---
## manage-user-feedback-settings-on-claude-console-016ac7151e.md

# Manage user feedback settings on Claude Console

March 16, 2026

Copy for LLM

As an Admin of your Claude Console account, you can manage the ability for members of your organization to submit feedback to Anthropic via thumbs up / thumbs down.

To manage feedback for your Console organization:

1. Navigate to [Settings \> Privacy controls](https://platform.claude.com/settings/privacy).

2. Toggle the feedback switch on or off.

More information on how Anthropic collects, uses, and stores feedback data can be found in our Privacy Center: [How long do you store my organization’s data?](https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data)

---
## manage-user-feedback-settings-on-claude-console.md

4. Manage user feedback settings on Claude Console

# Manage user feedback settings on Claude Console

As an Admin of your Claude Console account, you can manage the ability for members of your organization to submit feedback to Anthropic via thumbs up / thumbs down.

To manage feedback for your Console organization:

1. Navigate to [Settings \> Privacy controls](https://platform.claude.com/settings/privacy).

2. Toggle the feedback switch on or off.

---
## model-safety-bug-bounty-program-d990265deb.md

# Model Safety Bug Bounty Program

Copy for LLM

## Purpose

We believe external testing is crucial for building a safe AI ecosystem. As model capabilities advance, the consequences of jailbreaks could become increasingly significant. This ongoing Program builds on our successful previous bug bounty initiatives with several key objectives:

1. Identify universal jailbreaks in our deployed systems with ASL-3 protections

2. Provide continuous assessment of our safeguards' effectiveness

3. Test our monitoring systems' ability to detect vulnerabilities

4. Incentivize the user of one central legitimate channel for reporting publicly available jailbreaks

## Program Overview

Our Model Safety Bug Bounty Program is run through HackerOne. Through this Program, we are interested in finding universal jailbreaks that surpass our [Constitutional Classifiers](https://www.anthropic.com/news/constitutional-classifiers) system. We also occasionally run targeted programs within our overall Program to test the robustness of classifiers we hope to launch in the future.

A universal jailbreak is a generalized technique that reliably elicits policy-violating responses from a language model, regardless of the input prompt. Unlike narrow jailbreaks, which depend on the specifics of a particular question or context, universal jailbreaks work across a wide range of prompts and scenarios.

This is an ongoing Program. Once accepted to the Program on HackerOne, participants can submit jailbreak reports at any time through this Program. **To help with your red-teaming efforts, we provide access to a free model alias that reflects the model and classifiers live on our latest, most advanced model**. **Your use of this free model alias must be limited to performing authorized red-teaming activities.**

## Program Scope

This Program is primarily interested in discovering jailbreaks that are **universal**, in that they can reveal harmful information across a wide range of queries, and **detailed**, in that they reveal highly specific harmful information related to biological threats.

To emphasize, we are interested in jailbreaks that extract information that answers a set of **harmful biological questions that we share with accepted participants in the Program.**

We will pay **up to \$35,000 per novel, universal jailbreak identified.** We are only interested in jailbreaks that reveal substantial amounts of harmful information based on our sole criteria and discretion. We award bounties using a sliding scale based on an internal grading rubric which determines how detailed and accurate responses are.

This program is scoped to jailbreaks on our Constitutional Classifiers. For technical vulnerabilities that potentially exist on our Information Systems such as misconfigurations, CSRFs or cross site request forgeries, privilege escalation attacks, SQL Injection, XSS, and directory traversal attacks, please refer to our [Responsible Disclosure Policy](https://www.anthropic.com/responsible-disclosure-policy) and submit your report [here](https://hackerone.com/4f1f16ba-10d3-4d09-9ecc-c721aad90f24/embedded_submissions/new).

## How to Apply

You can apply to join our Program [here](https://docs.google.com/forms/d/e/1FAIpQLSf3IuyunFH1Rbz_9Bpt2kGBfwSW5QQ1TBkeAzNZrtCP-hRvNA/viewform). We review applications on a rolling basis. If accepted, you will receive an invite via HackerOne. IIf you do not already have a HackerOne account, please create one before applying to the Program so we can invite you directly on the platform. You must use your @wearehackerone.com email alias to create a [Claude Console account](https://platform.claude.com).

## Disclosure Guidelines & Confidentiality Obligations

All Program participants are required to sign a non-disclosure agreement to protect Program confidentiality as a condition for joining. **You may publicly disclose:**

- The existence of Anthropic's Model Safety Bug Bounty Program.

- Your participation as a selected participant in the Program.

**You may not disclose without express permission:**

- Any jailbreaks/vulnerabilities (even resolved ones) outside of the Program without express consent from Anthropic.

- The testing question set.

- Details about the classifiers and safety mitigations.

- Information about the models being tested.

- Identity of other participants.

- Any other information related to the Program, except as expressly permitted above.

## Anthropic’s Use of Data from the Program

Participant agrees that all data submitted to Anthropic, including its products and services, in connection with this Program may be used, stored, shared, and/or published by Anthropic indefinitely in furtherance of its safety research, model development, and related purposes without further obligation to Participant.

---
## model-safety-bug-bounty-program.md

# Model Safety Bug Bounty Program

## Purpose

1. Identify universal jailbreaks in our deployed systems with ASL-3 protections

2. Provide continuous assessment of our safeguards' effectiveness

3. Test our monitoring systems' ability to detect vulnerabilities

4. Incentivize the user of one central legitimate channel for reporting publicly available jailbreaks

## Program Overview

## Program Scope

To emphasize, we are interested in jailbreaks that extract information that answers a set of **harmful biological questions that we share with accepted participants in the Program.**

This program is scoped to jailbreaks on our Constitutional Classifiers. For technical vulnerabilities that potentially exist on our Information Systems such as misconfigurations, CSRFs or cross site request forgeries, privilege escalation attacks, SQL Injection, XSS, and directory traversal attacks, please refer to our [Responsible Disclosure Policy](https://www.anthropic.com/responsible-disclosure-policy) and submit your report [here](https://hackerone.com/297a385f-b3bd-4ecd-9466-7d9ad55371ce/embedded_submissions/new).

## How to Apply

## Disclosure Guidelines & Confidentiality Obligations

All Program participants are required to sign a non-disclosure agreement to protect Program confidentiality as a condition for joining. **You may publicly disclose:**

- The existence of Anthropic's Model Safety Bug Bounty Program.

- Your participation as a selected participant in the Program.

**You may not disclose without express permission:**

- Any jailbreaks/vulnerabilities (even resolved ones) outside of the Program without express consent from Anthropic.

- The testing question set.

- Details about the classifiers and safety mitigations.

- Information about the models being tested.

- Identity of other participants.

- Any other information related to the Program, except as expressly permitted above.

## Anthropic’s Use of Data from the Program

---
## models-usage-and-limits-in-claude-code-8e87d4473b.md

# Models, usage, and limits in Claude Code

Copy for LLM

This guide explains which model you are using, how usage is metered, and how to keep long sessions within their context and usage limits.

------------------------------------------------------------------------

## How usage is metered

How you signed in determines how usage is metered. Everything else about Claude Code behaves the same way regardless.

[TABLE]

If you signed in with an Enterprise seat, you generally do not need to think about tokens until you reach a limit. If you are using an API key, the **`/cost`** command shows your running spend for the current session.

------------------------------------------------------------------------

## Choosing a model

Run **`/model`** at any time to see which model you're using, check which models are available to your account, and switch between them. Your default model depends on how you signed in and on your organization's settings. As a rough guide:

- **Sonnet** is the right choice for the large majority of coding work. It is fast, capable, and cost-efficient.

- **Opus** offers deeper reasoning for harder problems such as large cross-cutting refactors, difficult debugging, or architectural decisions. It uses meaningfully more of your quota, so consider switching to Sonnet for routine work.

- **Haiku** is the fastest and cheapest option, well suited to quick lookups, simple edits, or high-volume scripted runs.

You can change models mid-session without losing your conversation. A common pattern is to plan with Opus and execute with Sonnet.

**Note:** Exact model names, versions, and availability change over time. The `/model` command is always the source of truth for your account.

------------------------------------------------------------------------

## What actually consumes tokens

Every turn sends three things to the model:

1. **The conversation so far** — every previous message in this session.

2. **Project context** — your `CLAUDE.md` and any files Claude has read.

3. **Your new prompt.**

Of these, the first item grows the fastest. A long debugging session in which Claude has read twenty files and produced fifteen diffs is carrying all of that on every subsequent message. This is where both cost and context limits originate.

------------------------------------------------------------------------

## Managing the context window

The **context window** is the maximum amount of text the model can consider at once. Claude Code shows a live indicator of how full it is. When it fills up, Claude can no longer see the oldest parts of the conversation clearly and quality drops.

Two commands keep it under control:

- **`/clear`** wipes the conversation and starts fresh. Your `CLAUDE.md` and project files remain available; only the chat history is removed. Use this whenever you switch tasks, as it is the single most effective lever for both quality and cost.

- **`/compact`** summarizes the conversation so far into a short recap, freeing up space while preserving the essential context. Use this when you are mid-task and need to keep going. Claude Code also auto-compacts when you get close to the limit, so you will rarely hit a hard wall.

**Rule of thumb:** use `/clear` when starting a new task, and `/compact` when continuing a long one.

------------------------------------------------------------------------

## Five habits that stretch your usage furthest

Almost every "I burned through my limit by lunchtime" report traces back to one of these five.

### 1. Clear between tasks

Every previous message is resent on every turn, so a session that has wandered through three unrelated problems pays for all three on each new message.

**In practice:** you just finished debugging a login redirect and now want to write a database migration. Run `/clear` first. A simple test: if your next prompt would make perfect sense in a brand-new terminal, clear before sending it. Your `CLAUDE.md` and project files stay put; only the chat history goes.

**One warning:** `/clear` cannot be undone. If you might still need something from the history, copy it out first or run `/compact` instead, which preserves a summary rather than wiping everything.

### 2. Match the model to the job

Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku. Spending Opus on routine work is the fastest way to drain a daily limit.

**Reasonable defaults:** Sonnet for most coding (features, tests, known bugs, refactors); Opus when you're genuinely stuck or the change is wide (hard debugging, cross-cutting refactors, architecture calls); Haiku for quick mechanical work (renames, log lines, regex explanations, boilerplate).

### 3. Point at files instead of pasting them

Anything you paste sits in context, in full, for the rest of the session. Referencing a file by path lets Claude read selectively and focus on the part you care about.

**In practice:** instead of pasting `auth.ts`, write look at the `validateToken` function in `src/auth.ts` — mentioning the path lets Claude open and read selectively. (Note that the @ prefix injects the entire file plus its `CLAUDE.md` tree into context, so use a bare path when you are trying to *save* tokens.) For logs and stack traces, trim to the relevant 20 or 30 lines before pasting. For anything large (lockfiles, build logs, data dumps), put it on disk and reference the path.

### 4. Keep **CLAUDE.md** lean

This file is prepended to *every* turn. Prompt caching means turns after the first are billed at the much cheaper cache-read rate, so the dollar cost is lower than the raw line count suggests, but it still occupies context-window space on every message.

**The rule: two strikes, keep it tight.** Only add a note the second time you have to correct Claude on the same thing (first-time issues are usually one-offs). Keep the file under roughly 200 lines; if something new needs to go in and there's no room, something old has to come out.

**When to update it:** right after a session where you had to correct Claude twice on the same thing. That's when the fix is fresh and takes a minute to write down. Every few weeks, read the whole file and delete anything that is no longer true or whose purpose you can't remember. Stale notes are worse than missing notes because they actively misdirect Claude.

### 5. Ask for a plan before big changes

A plan costs a few hundred tokens. A wrong 400-line diff that you revert and regenerate costs thousands, twice, plus the turns spent explaining what went wrong. **In practice:** for anything touching more than two or three files, switch to Plan Mode or just ask: "Before changing anything, list the files you'll touch and what you'll do in each." Read the list, correct it in plain English ("skip `legacy/`, and don't touch the tests yet"), then let it execute.

**Pro tip: plan with Opus, execute with Sonnet.** The highest-value use of Opus is writing the plan itself, where deeper reasoning actually pays off. Once a good plan exists, execution is mostly mechanical and Sonnet handles it at a fraction of the cost. This pattern is built in as `/model opusplan`, which uses Opus while planning and Sonnet for execution. Switching models doesn't clear the conversation, so Sonnet still sees everything Opus produced.

------------------------------------------------------------------------

## What to do when you reach a limit

- **Enterprise seat users:** the message tells you when your window resets. In the meantime you can switch to a lighter model with `/model`, or, if your organization allows it, temporarily fall back to an API key.

- **API key users:** there is no usage cap, but check `/cost` and your Console or cloud-provider dashboard if spend is a concern. Unexpectedly high numbers almost always trace back to very long sessions that were never cleared.

- **Context window full** (which is different from a usage limit): run `/compact` to keep going, or `/clear` if the older history is no longer needed.

------------------------------------------------------------------------

## Quick reference

[TABLE]

---
## online-safety-contacts-8adb6d08ad.md

# Online Safety Contacts

March 16, 2026

Copy for LLM

# **Designated Point of Contact for Authorities**

Government authorities can contact Anthropic directly by emailing [\[email protected\]](/cdn-cgi/l/email-protection#b7c5d2d0c2dbd6c3d8c59aded9c6c2dec5ded2c4f7d6d9c3dfc5d8c7ded499d4d8da). This is reserved for government officials seeking a point of contact and for governmental submissions to remove content.

For other types of inquiries, please use the mechanisms described below.

# **Illegal Content Reporting**

For user requests for content removal, please use the mechanisms described [here](https://support.anthropic.com/en/articles/10684638-blocking-and-removing-content-from-claude).

---
## online-safety-contacts.md

# Online Safety Contacts

# **Designated Point of Contact for Authorities**

Government authorities can contact Anthropic directly by emailing [\[email protected\]](/cdn-cgi/l/email-protection#8dffe8eaf8e1ecf9e2ffa0e4e3fcf8e4ffe4e8fecdece3f9e5ffe2fde4eea3eee2e0). This is reserved for government officials seeking a point of contact and for governmental submissions to remove content.

For other types of inquiries, please use the mechanisms described below.

# **Illegal Content Reporting**

For user requests for content removal, please use the mechanisms described [here](https://support.anthropic.com/en/articles/10684638-blocking-and-removing-content-from-claude).

---
## our-approach-to-user-safety-f95ca008dd.md

# Our Approach to User Safety

March 16, 2026

Copy for LLM

User safety is core to Anthropic’s mission of creating reliable, interpretable, and steerable AI systems. As we launch new ways for people to interact with Claude, we also expect to see new types of potential harm materialize, whether through the generation of misinformation, objectionable content, hate speech or other misuses. We are actively investing in and experimenting with additional safety features to supplement our existing model safety efforts and are working to provide helpful tools to a wide audience while also doing our best to mitigate harm. Launching new products in open beta allows us to experiment, iterate and hear your feedback. Here are some of the safety features we’ve introduced:

- Detection models that flag potentially harmful content based on our [Usage Policy](https://vault.pactsafe.io/s/9f502c93-cb5c-4571-b205-1e479da61794/legal.html#aup).

- Safety filters on prompts, which may block responses from the model when our detection models flag content as harmful.

- Enhanced safety filters, which allow us to increase the sensitivity of our detection models. We may temporarily apply enhanced safety filters to users who repeatedly violate our policies, and remove these controls after a period of no or few violations.

These features are not failsafe, and we may make mistakes through false positives or false negatives. Your feedback on these measures and how we explain them to users will play a key role in helping us improve these safety systems, and we encourage you to reach out to us at [\[email protected\]](/cdn-cgi/l/email-protection#d8adabbdaaabb9bebdaca198b9b6acb0aab7a8b1bbf6bbb7b5) with any feedback you may have. To learn more, [read about our core views on AI safety](https://www.anthropic.com/index/core-views-on-ai-safety).

---
## our-approach-to-user-safety.md

# Our Approach to User Safety

- Detection models that flag potentially harmful content based on our [Usage Policy](https://vault.pactsafe.io/s/9f502c93-cb5c-4571-b205-1e479da61794/legal.html#aup).

- Safety filters on prompts, which may block responses from the model when our detection models flag content as harmful.

These features are not failsafe, and we may make mistakes through false positives or false negatives. Your feedback on these measures and how we explain them to users will play a key role in helping us improve these safety systems, and we encourage you to reach out to us at [\[email protected\]](/cdn-cgi/l/email-protection#9eebedfbecedfff8fbeae7defff0eaf6ecf1eef7fdb0fdf1f3) with any feedback you may have. To learn more, [read about our core views on AI safety](https://www.anthropic.com/index/core-views-on-ai-safety).

---
## preserved-thinking-changing-how-the-messages-api-handles-thinking-blocks-to-prot-c453b32a3c.md

# Preserved thinking: changing how the Messages API handles thinking blocks to protect against distillation

Updated today

Copy for LLM

We're changing how the Messages API handles thinking blocks to protect against distillation. A thinking block is a record of the reasoning Claude may produce while working on a response. On Claude Fable 5.1, Claude Opus 5.5, and Claude Sonnet 5.5, new API accounts can no longer edit the context around a thinking block, such as the messages, tools, or system prompt, during a multi-turn conversation. We’ll expand the rollout to all users with upcoming model launches.

Modifying this prior context has legitimate applications, which we continue to support using the adjustments outlined below. However, such modifications are also a common and **[publicly documented technique](https://arxiv.org/abs/2608.09867)** for industrial-scale illicit distillation, which is prohibited by our **[Usage Policy](https://www.anthropic.com/legal/aup)** and Terms of Service.

Now, the API will now verify that a thinking block is sent back with the same system prompt, tools, and messages that produced it, and will return an error if they don't match. In order for modified requests to succeed, developers may opt-in to instead have the thinking blocks *removed* from such requests; the model will respond without seeing the thinking block.

In this article, we share details on why we're doing this and the adjustments you can make to minimize disruption.

## What are thinking blocks?

Claude produces reasoning steps before providing its final answer. On the API, these are returned to the user as "thinking blocks." In a multi-turn conversation, API users send these blocks back with each exchange (along with the system prompt, tools, and earlier messages), so that Claude has full conversational context.

## What’s changing?

For affected accounts using the models listed below, the API will return an error if the system prompt, tools, or messages preceding a prior thinking block have been modified.

To avoid an error message, you may opt into "non-strict" mode. In this mode, the request will go through, but the affected thinking blocks will be dropped from what the model sees. This allows you to continue your conversation or task uninterrupted despite the prior turns’ thinking not being shown to the model. When this happens, the API response will tell you which blocks were dropped.

## Why are we making this change?

Altering the earlier turns of a conversation is a **[common technique](https://arxiv.org/abs/2608.09867)** used in illicit distillation campaigns, which aim to extract the capabilities of advanced models—especially thinking—to train another model, without authorization. Distillation is often employed on an industrial scale, using thousands of fake accounts. We encrypt Claude's thinking blocks to prevent this, but by editing the conversation before a thinking block, a user could get Claude to decrypt and print its reasoning. Systems trained this way can inherit capabilities they wouldn’t otherwise have, *without* inheriting the safeguards that we've built to prevent a broad range of misuse like cyberattacks and weapons development.

This change aims to make distillation campaigns more difficult to execute. It builds on existing anti-distillation measures like **[enhanced distillation classifiers](https://www.anthropic.com/research/next-generation-constitutional-classifiers)** and restrictions on transferring sessions or reasoning from more advanced models to less capable models with weaker safeguards.

## What does this mean for API integrations?

Certain integrations—particularly those that involve rewriting earlier turns mid-conversation, like context compaction, injected system reminders, or changing tools mid-session—may need adjustment.

Here are the resources to help guide you through this update:

- The **[preserved thinking documentation](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking)** covers where the change applies, how to check whether your integration is affected, and **[how to make common edits](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#replace-prefix-edits)** without breaking preserved thinking.

- **[Compaction](https://platform.claude.com/docs/en/build-with-claude/compaction)** (beta) can summarize older turns while keeping the most recent turns word for word, and can build the summary in the background while your agent keeps working. It replaces client-side compaction, which is likely the most common reason integrations need to change.

- **[Mid-conversation tool changes](https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages#mid-conversation-tool-changes)** (beta) let you add or remove tools during a conversation without editing earlier turns.

- Our **[migration guides](https://platform.claude.com/docs/en/about-claude/models/migration-guide)** have a full checklist for moving to each model.

Check each page for availability on Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

If the guidance above doesn't cover your use case, please **[reach out to our support team](https://support.claude.com/en/articles/9015913)**. If you work with an account team, you can also reach out to them for support with updating more complex integrations.

There are additional benefits to keeping thinking blocks consistent: it means that the API can reuse cached prompts more often, which reduces costs and response time.

## Who will this impact?

Preserved thinking applies to these models and accounts:

[TABLE]

This covers Claude Platform organizations, Amazon Bedrock accounts, Google Cloud Vertex AI projects, and Microsoft Foundry projects.

We're taking a phased approach to enforcement, starting with new accounts, where we see the highest concentration of distillation-related abuse. This gives developers with existing accounts time to make their harnesses and integrations compatible.

If you use Claude Code, Claude Cowork, or Claude.ai, there's nothing you need to change; those products handle thinking blocks for you.

## Thinking can only be read by the account that created it

Starting with Claude Sonnet 5.5, a thinking block can only be read by the account that created it, or by an account linked to it.

Thinking blocks can contain private information. Because they are encrypted, that information isn't visible when you inspect a transcript. This check helps keep a shared or leaked transcript from exposing the reasoning inside it. It also makes it harder for distillers to move harvested transcripts to new accounts after we ban the account that produced them.

If a request includes thinking from an account that isn't linked, the API drops that thinking and the request continues. It doesn't return an error. The next response may be slower and use more tokens, similar to a compaction.

Accounts in the same Claude Platform parent organization or the same Google Cloud organization are linked automatically. If you continue conversations across other accounts, for example to fail over between the Claude Platform and Amazon Bedrock, contact your account team to link them.

In Claude Code, switching accounts in the middle of a session has the same effect: the next response is slower and uses more tokens.

---
## project-glasswing.md

# Project Glasswing: Securing Critical Software for the AI Era

Project Glasswing is Anthropic's cross-industry initiative to secure the world's most critical software, powered by Claude Mythos Preview — Anthropic's most capable AI model for cybersecurity tasks.

Announced on April 7, 2026.

## Overview

Project Glasswing partners the organizations responsible for infrastructure that billions of people depend on with Claude Mythos Preview, giving their defenders a head start in the coming AI-driven era of cybersecurity.

Anthropic is not making Claude Mythos Preview generally available due to its cybersecurity capabilities. Instead, access is restricted to vetted partners who build or maintain critical software infrastructure.

## Launch Partners

The initiative brings together the following organizations as launch partners:

- Amazon Web Services (AWS)
- Anthropic
- Apple
- Broadcom
- Cisco
- CrowdStrike
- Google
- JPMorganChase
- Linux Foundation
- Microsoft
- NVIDIA
- Palo Alto Networks

Anthropic has also extended access to a group of over 40 additional organizations that build or maintain critical software infrastructure.

## Financial Commitment

- **$100M** in Claude Mythos Preview usage credits committed across Project Glasswing participants throughout the research preview
- **$4M** in direct donations to open-source security organizations:
- $2.5M to Alpha-Omega and the Open Source Security Foundation through the Linux Foundation
- $1.5M to the Apache Software Foundation

## Claude Mythos Preview

Claude Mythos Preview is a new general-purpose language model that sits in a fourth model tier named **Copybara**. Anthropic describes it as superior to any other existing AI frontier model.

### Capabilities

- Capable of identifying and exploiting zero-day vulnerabilities in every major operating system and every major web browser
- Has autonomously discovered thousands of zero-day vulnerabilities during testing
- Can chain together 3, 4, or sometimes 5 vulnerabilities in sequence to produce sophisticated exploits
- Demonstrates "a level of coding capability where [it] can surpass all but the most skilled humans at finding and exploiting software vulnerabilities"

### Benchmarks

| Benchmark | Score |
|-----------|-------|
| SWE-bench Verified | 93.9% |
| GPQA Diamond | 94.6% |
| CyberGym | 83.1% |

Highest-scoring model on record across all three benchmarks.

### Availability

- **Not publicly available** — restricted to Project Glasswing partners and vetted organizations
- Available through the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry for participants
- Pricing described as "very expensive" but specific per-token pricing not publicly disclosed

## Why Limited Access

As Newton Cheng, Frontier Red Team Cyber Lead at Anthropic, explained: Anthropic is not making the model generally available due to concerns that it may pose unprecedented cybersecurity risks and could increase the likelihood of large-scale AI-driven cyberattacks. The restricted access is designed to give defenders a head start before the capabilities become more widely available.

## See Also

- [red.anthropic.com](red-anthropic-com-index.md) — Anthropic's cybersecurity research site
- [Claude Mythos Preview](../20-Models/claude-mythos-preview.md) — Technical details
- [Building AI for cyber defenders](../19-Reference/building-ai-for-cyber-defenders-anthropic.md)

---
## provision-and-manage-skills-for-your-organization-68c1434d53.md

# Provision and manage skills for your organization

Copy for LLM

This article explains how organization owners can provision skills for everyone in their organization, and how to scope skills to specific groups using plugins. Provisioning skills lets you distribute approved workflows and capabilities across your organization from one place.

Organization-wide skill management is available to Team and Enterprise plans.

## Prerequisites

Before you can provision skills for your organization, navigate to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**, select the “Policy” tab, and check that both **Cloud code execution and file creation** and **Skills** are toggled on. Skills require code execution to function, so if code execution is disabled, skills will not be available.

------------------------------------------------------------------------

## Provision skills for everyone

When you upload a skill through organization settings, it becomes available to everyone in your organization in **[Customize \> Skills](https://claude.ai/customize/skills)**. Individual users no longer need to upload the same skill themselves. Provisioned skills also load in Claude Code for users who sign in with their Claude account. To stop skills from syncing to Claude Code, set `syncClaudeAiSkills` to `false` in Claude Code managed settings.

**To provision a skill:**

1. Navigate to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**.

2. Click "Add” in the upper right corner.

3. Choose “Upload a skill” or “Create a skill.”

4. If uploading, select a .zip file containing your skill (must include a SKILL.md file).

5. If creating, input the skill name and description in the modal, then click “Create.”

6. The skill is immediately provisioned to all users in your organization.

Owner-provisioned skills are enabled by default for everyone, but users can toggle individual skills off if they choose. This gives your organization consistent, approved workflows while letting users customize their own experience.

------------------------------------------------------------------------

## Provision skills to specific groups

Provisioning a skill through **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** gives it to everyone. Enterprise plans can give skills to only some users by bundling them into a plugin and assigning that plugin to a group. The group's members see those skills, and members outside the group don't.

For example, if you have 10 skills for your marketing team, add them to a plugin and assign it to the marketing group. Only that group gets those skills.

Skills provisioned this way appear in chat (on the web and the "Chat" tab in Claude Desktop), in Claude Cowork, and in Claude Code in the terminal for users who sign in with their Claude account. Group targeting you've already set up for Cowork carries over to chat with no extra steps.

To set this up, see **[Manage plugins for your organization](https://support.claude.com/en/articles/13837433)**.

------------------------------------------------------------------------

## Control whether users can create skills

By default, users can create their own skills in Claude and upload skill files to their personal skills list. If you'd rather users only use skills you've provisioned, you can turn this off for your organization.

To turn off skill creation for users:

1. Navigate to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and click the “Policy” tab.

2. Turn off **User-created skills**.

When **User-created skills** is off:

- Users can't create skills in Claude or upload skill files.

- Skills you've provisioned and Anthropic's built-in skills stay available, and users can still enable and use them.

**Note:** If your organization uses custom roles, a user also needs the **Create skills** capability on their role. The organization setting is the main switch: when it's off, users can't create or upload their own skills, regardless of role, but Owners can still provision skills for the organization. When it's on, Enterprise plan users on custom roles still need the role capability. Learn more about **[managing custom roles on Enterprise plans](https://support.claude.com/en/articles/13930452)**.

------------------------------------------------------------------------

## Let users publish skills and plugins to your organization

Users can submit a skill or plugin they've built to your organization's library so everyone can use it. When your organization requires review, an owner approves each submission before it's published, and every later version goes through the same review.

### Set the publishing policy

1. Navigate to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and select the “Policy” tab.

2. Under **Publishing**, choose an option:

- **Requires review:** Users can submit skills and plugins, and an owner must approve each one before it's published.When submitting, the user proposes how the item is offered. You can accept or change that when you approve.

- **Open:** Skills and plugins that users submit are published to the organization library without review, and are available to everyone in your organization to install. Users can't choose a different offering. On Enterprise plans with **[Skill and plugin security scanning](https://support.claude.com/en/articles/15927065)** on, a new item isn't listed for others until it passes the scan, which usually takes a few minutes.

- **Off:** Users don't see the "Publish to org" button. Owners can still add skills and plugins to the organization directly.

Your starting setting depends on your plan:

- **Team plans:** Publishing is set to “Open,” unless you already had **Share with organization** turned off, in which case “Off” is the default.

- **Enterprise plans:** Publishing starts off.

- If your organization already had **Share with organization** turned on, it starts as “Open” instead.

- If you haven't chosen a setting, it switches to “Requires review” on October 2, 2026. To keep publishing off, or to choose a different setting, select it yourself before then.

Turning **Publishing** off doesn't remove anything that's already published. Published items stay in the library until an owner removes them.

### Review a submission

Owners, and anyone whose role has Libraries set to “Can manage,” can review requests. You can't approve your own.

**To review a submission:**

1. Navigate to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and select the “Requests” tab.

- Plugin and skillsubmissions appear here, and each request shows who submitted it, whether it's a skill or plugin, its security scan result, and its status. If a submission looks like a skill or plugin your organization already has, a **Possible duplicate** tag appears next to it.

2. Open a request. You'll see:

- How the item will be offered

- The scan result

- Every file

- For an update, what changed since the currently published version

- Up to three possible duplicates, if any, each with a reason and a link

3. Choose how the item is offered and who gets it: everyone or specific groups. The submitter's proposed offering is preselected, and you can change it. Then click "Approve" to publish it, or click "Request changes" and add a note for the person who submitted it.

Keep these points in mind when reviewing:

- **You review a fixed version.** You see exactly the version that was submitted. The author can keep editing their own copy, but those edits don't reach the organization until they submit a new version and it's approved.

- **Possible duplicate hints are a note, not a block.** Only reviewers see them. They don't stop you from approving a submission, and the person who submitted it doesn't see them. No setup is needed. Duplicate checks don't run for organizations with HIPAA, zero data retention, FERPA, or customer-managed keys.

- **For Enterprise plans, submitted versions are scanned.** For Enterprise plans with **Skill and plugin security scanning** enabled, each version is checked for malicious content automatically. A submission that fails the scan can't be approved until the author fixes it and submits again. Learn more about **[skill and plugin scanning](https://support.claude.com/en/articles/15927065)**.

### Manage published skills and plugins

Every approved item is listed on the “Inventory” tab in **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)**. To see only your organization's items, set the **Source** filter to “Your organization."

Click the menu button at the end of a row to:

- View details and files

- View version history

- Choose how the item is offered under Default access (Available to install, Installed by default, Not available, or Required)

- Set access for specific groups under “Group access…”

- Copy a link to share with users

The “Inventory” tab also lists skills and plugins that users created or shared.

When you approve a new version, everyone who uses the item gets the update automatically. Until then, users stay on the currently approved version.

Published items are managed by your organization. If the author leaves, the published item stays in the library.

------------------------------------------------------------------------

## Control skill sharing between users

In addition to provisioning skills top-down, you can let users share skills or plugins they've built with each other. Two independent toggles control this:

- **Skill sharing:** Users can share a skill or plugin with specific colleagues. Recipients see the skill in the **Shared with you** section of their skills list.

- **Share with groups:** Users can share a skill or plugin with an entire group. Recipients see it in the **Shared with you** section of their skills list, the same as items shared with individuals.

To let users add skills and plugins to the organization library, use the **Publishing** setting. Learn more about **[letting users publish skills and plugins to your organization](#h_1abc45a27c)**.

The **Skill sharing** toggle is on by default for Team plans and for Enterprise plans that haven't set a skills preference. For organizations with HIPAA readiness or other regulated configurations, skills and skill sharing are off by default and an admin can enable them in **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** in the “Policy” tab. The **Share with groups** toggle is also off by default and can be enabled by an admin.

**Note:** Shared skills and plugins are view-only, and stay off until the recipient chooses to enable them. Recipients can enable and use a shared skill or plugin but can't edit its contents. When the owner saves a new version, everyone it's shared with gets the update automatically at next use. The owner can revoke someone's access at any time, and access is removed automatically if the recipient leaves the organization.

### Share skills with a group

If you use custom roles, also make sure the **Share skills with groups** capability is enabled for their role. See **[Manage custom roles on Enterprise plans](https://support.claude.com/en/articles/13930452)**.

Once these settings are on, users can begin sharing skills with groups.

### How shared skills differ from provisioned skills

[TABLE]

**Important:** When Publishing is set to “Open,” anything a user publishes goes straight to the organization library without review and is available to everyone in your organization Choose “Requires review” if you want an owner to check each skill and plugin first.

### Monitor sharing activity

Skill sharing events are captured in the audit log and Compliance API as `role_assignment` events. You can see who shared a skill or plugin, with whom, and whether it was peer-to-peer, a group, or (skills only) organization-wide.

The audit log doesn't capture the contents of shared skills or plugins—only the share event itself. The “Inventory” tab shows metadata, sharing status, and scan status for skills and plugins that users created or shared, but not their contents.

------------------------------------------------------------------------

## How users see provisioned and shared skills

Skills appear for each user in **[Customize \> Skills](https://claude.ai/customize/skills)**, organized into three sections:

- **Personal skills:** Skills the user has created or uploaded.

- **Shared with you:** Skills colleagues have shared directly with a user. These appear grayed out until enabled.

- **Organization skills:** Skills an owner has provisioned and skills published to the organization. Users install these from the directory.

Owner-provisioned skills are marked with a visual indicator so users can distinguish them from other skill types. Users can click on any skill to preview its contents and description.

For more on how users browse and install from the directory, see **[Browse skills, connectors, and plugins in one directory](https://support.claude.com/en/articles/14328846-browse-skills-connectors-and-plugins-in-one-directory)**.

------------------------------------------------------------------------

## Manage and remove provisioned skills

The “Inventory” tab in **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** lists all skills provisioned for your organization. Set **Source** to “Your organization” and **Type** to “Skills,” or search by name. To remove a skill, click the menu button at the end of its row and select “Remove from library.” Once removed, the skill will no longer appear in users' skills lists in **[Customize \> Skills](https://claude.ai/customize/skills).**

**Note:** Only owners can add or remove organization-wide skills. Individual users cannot delete provisioned skills, though they can toggle them off for their own use.

------------------------------------------------------------------------

## Scan skills and plugins for malicious content

On the Enterprise plan, you can turn on skill scanning for your organization. When it's on, Claude checks each third-party skill and plugin your users upload or edit for malicious content before it can run. Scanning is off by default until October 2, 2026. From that date it's on by default, where available, for Enterprise organizations that haven't set it. Organizations that already turned it on or off keep their choice.

To turn on skill scanning for your organization:

1. Go to **[Organization settings \> Plugins & skills](https://claude.ai/admin-settings/skills)** and select the “Policy” tab.

2. Turn on **Skill and plugin security scanning**.

If you use custom roles, you can further define who scanning applies to by turning on the **Skill and plugin security scanning** capability for roles that should have access to skill scanning.

Here's what your users see:

- A skill or plugin that passes the scan installs normally.

- A skill or plugin that may carry risk stays usable behind a caution banner the user acknowledges.

- A skill or plugin with malicious content is blocked and can't be used.

A blocked skill can't be overridden by the user who uploaded it, and can't be approved for the organization at this time. Scanning isn't available for organizations using customer-managed encryption keys (CMEK), zero data retention (ZDR), or HIPAA configurations. Learn more about **[skill and plugin scanning](https://support.claude.com/en/articles/15927065)**.

------------------------------------------------------------------------

## Best practices

- **Test skills before provisioning:** Upload and test skills on your own account first to verify they work as expected before distributing them organization-wide.

- **Scope specialized skills to groups:** When a skill is only relevant to one team, bundle it into a plugin and assign it to that group instead of provisioning it to everyone.

- **Use descriptive names:** Give skills clear names that help users understand their purpose at a glance.

- **Write clear descriptions:** The skill's description helps Claude determine when to use it automatically. Ensure descriptions accurately reflect what the skill does.

- **Consider default status carefully:** Enable skills by default when they're broadly useful to most users. Keep specialized skills disabled by default for the users who don't need them.

- **Decide on publishing deliberately:** Set Publishing to “Requires review” if you want an owner to check skills and plugins before they reach everyone.

---
## provision-and-manage-skills-for-your-organization.md

4. Provision and manage Skills for your organization

# Provision and manage Skills for your organization

Organization-wide skill management is available to Team and Enterprise plans.

This article explains how organization owners can provision skills for all users in their organization. Provisioning skills to your team allows you to distribute approved workflows and capabilities across your entire organization from a central location.

## Prerequisites

Before you can provision skills for your organization, you must enable two capabilities by toggling them on:

1. **Code execution and file creation** in **[Organization settings \> Capabilities](https://claude.ai/admin-settings/capabilities)**

2. **Skills** in **[Organization settings \> Skills](https://claude.ai/admin-settings/skills)**

Skills require code execution to function, so if code execution is disabled, skills will not be available.

------------------------------------------------------------------------

## Provision skills for your organization

When you upload a skill through organization settings, it automatically becomes available to all users in your organization in **[Customize \> Skills](https://claude.ai/customize/skills)**. This means that individual users no longer need to manually upload the same skills.

**To provision a skill:**

1. Navigate to **[Organization settings \> Skills](https://claude.ai/admin-settings/skills)**.

2. In the **Organization skills** section, click "+ Add."

3. Select a .zip file containing your skill (must include a SKILL.md file).

4. The skill is immediately provisioned to all users in your organization.

Admin-provisioned skills are enabled by default for everyone in your organization, but users can still toggle individual skills off if they choose. This gives organizations consistent, approved workflows across teams while letting individual users customize their experience.

------------------------------------------------------------------------

## How users see provisioned skills

Skills provisioned by an organization owner will appear for each user in **[Customize \> Skills](https://claude.ai/customize/skills)** alongside Anthropic skills and any skills they've uploaded themselves.

These skills are marked with a visual indicator so users can distinguish them from other skill types. Users can click on any skill to preview its contents and description.

------------------------------------------------------------------------

## Manage and remove provisioned skills

The **Organization skills** section in **[Organization settings \> Skills](https://claude.ai/admin-settings/skills)** displays all skills that have been provisioned for your organization. You can use search and section headings to navigate your provisioned skills.

To remove a skill from your organization, locate it in the **Organization skills** list and select the option to remove it. Once removed, the skill will no longer appear in users' Skills lists in **[Customize \> Skills](https://claude.ai/customize/skills).**

**Note:** Only owners can add or remove organization-wide skills. Individual users cannot delete provisioned skills, though they can toggle them off for their own use.

------------------------------------------------------------------------

## Best practices

- **Test skills before provisioning:** Upload and test skills on your own account first to verify they work as expected before distributing them organization-wide.

- **Use descriptive names:** Give skills clear names that help users understand their purpose at a glance.

- **Write clear descriptions:** The skill's description helps Claude determine when to use it automatically. Ensure descriptions accurately reflect what the skill does.

- **Consider default status carefully:** Enable skills by default when they're broadly useful to most users. Use disabled by default for specialized skills that only some team members need.

---
## red-anthropic-com-index.md

# red.anthropic.com — Anthropic Cybersecurity Research

red.anthropic.com is Anthropic's dedicated cybersecurity research site, publishing technical analyses of Claude's capabilities in vulnerability discovery, exploitation, and defensive security.

## Articles

### 2026

| Date | Title | Topic |
|------|-------|-------|
| Apr 8, 2026 | [Claude Mythos Preview](../20-Models/claude-mythos-preview.md) | New model with step-change cybersecurity capabilities |
| Mar 6, 2026 | [Partnering with Mozilla to improve Firefox's security](red-firefox-vulnerabilities.md) | Claude Opus 4.6 discovers 22 Firefox vulnerabilities |
| Mar 6, 2026 | [Reverse engineering Claude's CVE-2026-2796 exploit](https://red.anthropic.com/2026/exploit/) | Deep dive into a Firefox JIT miscompilation exploit |
| Feb 5, 2026 | [0-Days](https://red.anthropic.com/2026/zero-days/) | Claude Opus 4.6 finding vulnerabilities at scale |
| Jan 14, 2026 | [Property-Based Testing with Claude](https://red.anthropic.com/2026/property-based-testing/) | AI agents for property-based testing in Python |
| Jan 8, 2026 | [AI for Critical Infrastructure Defense](red-critical-infrastructure-defense.md) | PNNL partnership, water treatment plant simulation |
| 2026 | [AI Models on Realistic Cyber Ranges](red-cyber-toolkits-update.md) | Multistage attacks using only standard open-source tools |

### 2025

| Date | Title | Topic |
|------|-------|-------|
| Dec 18, 2025 | [Project Vend Phase Two](https://red.anthropic.com/2025/project-vend-2/) | AI managing vending machines and business operations |
| Nov 2025 | [AI for Cyber Defenders](https://red.anthropic.com/2025/ai-for-cyber-defenders/) | Improving Claude's vulnerability detection for defense |
| 2025 | [Smart Contracts](https://red.anthropic.com/2025/smart-contracts/) | AI agents find $4.6M in blockchain smart contract exploits |

---
## red-critical-infrastructure-defense.md

# AI for Critical Infrastructure Defense

Anthropic partnered with Pacific Northwest National Laboratory (PNNL) to explore AI's defensive application in protecting critical infrastructure from cyber attacks.

Published: January 8, 2026

## The ALOHA System

PNNL researchers created an adaptive generative AI agent called **ALOHA** (Agentic LLMs for Offensive Heuristic Automation) using Claude. The system demonstrated that AI can dramatically accelerate defensive red teaming exercises.

### How It Works

- PNNL developed a "scaffold" for Claude to automate and accelerate adversary emulation
- The scaffold translates natural language prompts into complex attack chains
- Pre-defined code-based "tools" allow the model to take actions on computer networks
- PNNL's estimate: attack reconstruction completed in **3 hours** instead of **multiple weeks**

## Water Treatment Plant Simulation

Using Claude, PNNL researchers emulated cyber attacks on a high-fidelity simulation of a water treatment plant in far less time than a human expert would have taken. This served as a proof of concept for how AI can help cyber defenders iterate faster on red teaming exercises.

## Partnership Benefits

- Laboratory researchers benefit from an advanced LLM for defensive research
- Anthropic subjects its technology to rigorous testing to prevent misuse
- The work demonstrates practical defensive applications of frontier AI capabilities

## See Also

- [AI for Cyber Defenders](https://red.anthropic.com/2025/ai-for-cyber-defenders/)
- [AI Models on Realistic Cyber Ranges](red-cyber-toolkits-update.md)

---
## red-cyber-toolkits-update.md

# AI Models on Realistic Cyber Ranges

Current Claude models can now succeed at multistage attacks on networks with dozens of hosts using only standard, open-source tools, instead of the custom tools needed by previous generations.

Published: 2026

## Key Findings

### Claude Sonnet 4.5 Progress

- Can succeed on a minority of test networks **without the custom cyber toolkit** needed by previous generations
- Can exfiltrate all simulated personal information in a high-fidelity simulation of the **Equifax data breach** using only a Bash shell on a widely-available Kali Linux host
- Uses standard, open-source penetration testing tools rather than a custom toolkit

### Rapid Improvement

- **Claude Sonnet 3.5** (released ~1 year before Sonnet 4.5): Could not succeed at the Equifax simulation in any of five trials without the specialized cyber toolkit
- **Claude Sonnet 4.5**: Succeeds using only publicly available tools

### Implications

The trajectory of models first needing specialized tools and then being able to operate without them (or using only publicly available tools) is consistent with other trends observed in AI progress. Anthropic believes it presages further improvement in the cyber domain.

## See Also

- [AI for Critical Infrastructure Defense](red-critical-infrastructure-defense.md)
- [Claude Mythos Preview](../20-Models/claude-mythos-preview.md)

---
## red-firefox-vulnerabilities.md

# Partnering with Mozilla to Improve Firefox's Security

In a collaboration with researchers at Mozilla, Claude Opus 4.6 discovered 22 vulnerabilities in Firefox over the course of two weeks.

Published: March 6, 2026

## Key Findings

- **22 CVEs discovered** in Firefox by Claude Opus 4.6
- **14 flagged as high severity** by Mozilla
- Vulnerabilities were fixed in **Firefox version 148** (released February 24, 2026)
- Bug reports included minimal test cases that allowed Mozilla's security team to quickly verify and reproduce each issue

## CVE-2026-2796: JIT Miscompilation Exploit

The most notable finding was CVE-2026-2796, a JIT miscompilation error in Firefox's JavaScript WebAssembly component:

- Assigned a **critical 9.8 CVSS score**
- Claude wrote an exploit achieving arbitrary read/write and code execution in Firefox's JavaScript engine by leveraging type confusion
- **Important limitation**: The exploit only works within a testing environment that intentionally removes some browser security features
- Claude produced a "crude" exploit in only 2 out of several hundred attempts, costing approximately $4,000 in API credits

## Reception

The collaboration was well-received by Mozilla. The Mozilla Blog post "[Hardening Firefox with Anthropic's Red Team](https://blog.mozilla.org/en/firefox/hardening-firefox-anthropic-red-team/)" describes the partnership from Mozilla's perspective.

## See Also

- [Reverse engineering Claude's CVE-2026-2796 exploit](https://red.anthropic.com/2026/exploit/)
- [0-Days: Vulnerability detection at scale](https://red.anthropic.com/2026/zero-days/)

---
## safeguards-warnings-and-appeals.md

# Safeguards Warnings and Appeals

Updated yesterday

*Please Note: Our response times are currently longer than normal due to our recent launch and an increase in email volume. We will reply to your appeal/email as soon as we can and we ask for your patience in the meantime.*

**Appeals**

As part of our safety process, we may ban an account for a variety of reasons:

- Repeated Violations of Our [Usage Policy](https://www.anthropic.com/legal/aup)

- Account Creation from an [Unsupported Location](https://support.anthropic.com/en/articles/8461763-where-can-i-access-claude-ai)

- [Terms of Service](https://www.anthropic.com/legal/consumer-terms) Violations

If you feel your account has been wrongly suspended or terminated, please fill out the appeal form linked below with your account information so our Safeguards team can further investigate why your account was disabled.

[Submit an appeal](https://docs.google.com/forms/d/e/1FAIpQLScs9d-gS4ivLuqfl3FKAgVZUkfX4ABONPH3d4fr3JmeRgjDdw/viewform)

**Cyber Restrictions**

As part of our safety process, we may apply restrictions to accounts that contain Cyber violative traffic. These restrictions are put in place to ensure Claude is used safely and in accordance with our [Usage Policy](https://www.anthropic.com/legal/aup).

If you work in cybersecurity and believe your account has been incorrectly restricted, please fill out the form below to share more context about your use case. Our Safeguards team will evaluate your submission and may adjust restrictions based on the additional information provided.

[Cyber Use Case Form](https://form.typeform.com/to/UGxlBd1z)

**Warnings**

As part of our safety process, we warn users if we believe their prompts are violating our [Usage Policy](https://www.anthropic.com/legal/aup). For API customers, these warnings are linked to ongoing thresholds of violative behavior across their entire API account.

If you believe we’ve made a mistake in issuing you a warning, please email [\[email protected\]](/cdn-cgi/l/email-protection#7e0b0d1b0c0d1f181b0a073e1f100a160c110e171d501d1113) with the details of your situation and your account information.

------------------------------------------------------------------------

Reporting, Blocking, and Removing Content from Claude

I’m planning to launch a product using the Claude API. What steps should I take to ensure I’m not violating Anthropic’s Usage Policy?

API Safeguards Tools

Exceptions to our Usage Policy

Reporting, Blocking, and Removing Content from Claude

---
## sandboxing-claude-code-docs-5f97cd27c4.md

# Sandboxing

Learn how Claude Code’s sandboxed bash tool provides filesystem and network isolation for safer, more autonomous agent execution.

> ## Documentation Index
>
> Fetch the complete documentation index at: <https://code.claude.com/docs/llms.txt>
>
> Use this file to discover all available pages before exploring further.

Overview

Claude Code features native sandboxing to provide a more secure environment for agent execution while reducing the need for constant permission prompts. Instead of asking permission for each bash command, sandboxing creates defined boundaries upfront where Claude Code can work more freely with reduced risk. The sandboxed bash tool uses OS-level primitives to enforce both filesystem and network isolation.

Why sandboxing matters

Traditional permission-based security requires constant user approval for bash commands. While this provides control, it can lead to:

- **Approval fatigue**: Repeatedly clicking “approve” can cause users to pay less attention to what they’re approving
- **Reduced productivity**: Constant interruptions slow down development workflows
- **Limited autonomy**: Claude Code cannot work as efficiently when waiting for approvals

Sandboxing addresses these challenges by:

1. **Defining clear boundaries**: Specify exactly which directories and network hosts Claude Code can access
2. **Reducing permission prompts**: Safe commands within the sandbox don’t require approval
3. **Maintaining security**: Attempts to access resources outside the sandbox trigger immediate notifications
4. **Enabling autonomy**: Claude Code can run more independently within defined limits

Effective sandboxing requires **both** filesystem and network isolation. Without network isolation, a compromised agent could exfiltrate sensitive files like SSH keys. Without filesystem isolation, a compromised agent could backdoor system resources to gain network access. When configuring sandboxing it is important to ensure that your configured settings do not create bypasses in these systems.

How it works

Filesystem isolation

The sandboxed bash tool restricts file system access to specific directories:

- **Default writes behavior**: Read and write access to the current working directory and its subdirectories
- **Default read behavior**: Read access to the entire computer, except certain denied directories
- **Blocked access**: Cannot modify files outside the current working directory without explicit permission
- **Configurable**: Define custom allowed and denied paths through settings

You can grant write access to additional paths using `sandbox.filesystem.allowWrite` in your settings. These restrictions are enforced at the OS level (Seatbelt on macOS, bubblewrap on Linux), so they apply to all subprocess commands, including tools like `kubectl`, `terraform`, and `npm`, not just Claude’s file tools.

Network isolation

Network access is controlled through a proxy server running outside the sandbox:

- **Domain restrictions**: Only approved domains can be accessed
- **User confirmation**: New domain requests trigger permission prompts (unless [`allowManagedDomainsOnly`](/docs/en/settings#sandbox-settings) is enabled, which blocks non-allowed domains automatically)
- **Custom proxy support**: Advanced users can implement custom rules on outgoing traffic
- **Comprehensive coverage**: Restrictions apply to all scripts, programs, and subprocesses spawned by commands

The built-in proxy enforces the allowlist based on the requested hostname and does not terminate or inspect TLS traffic. See [Security limitations](#security-limitations) for the implications of this design, and [Custom proxy configuration](#custom-proxy-configuration) if your threat model requires TLS inspection.

OS-level enforcement

The sandboxed bash tool leverages operating system security primitives:

- **macOS**: Uses Seatbelt for sandbox enforcement
- **Linux**: Uses [bubblewrap](https://github.com/containers/bubblewrap) for isolation
- **WSL2**: Uses bubblewrap, same as Linux

WSL1 is not supported because bubblewrap requires kernel features only available in WSL2. These OS-level restrictions ensure that all child processes spawned by Claude Code’s commands inherit the same security boundaries.

Getting started

Prerequisites

On **macOS**, sandboxing works out of the box using the built-in Seatbelt framework. On **Linux and WSL2**, install the required packages first:

- Ubuntu/Debian

- Fedora

```python
sudo apt-get install bubblewrap socat
```

```python
sudo dnf install bubblewrap socat
```

On Ubuntu 24.04 and later, the default AppArmor policy prevents bubblewrap from creating the user namespaces it needs for isolation. Add an AppArmor profile that grants `bwrap` this capability:

```python
sudo tee /etc/apparmor.d/bwrap > /dev/null <<'EOF'
abi <abi/4.0>,
include <tunables/global>

profile bwrap /usr/bin/bwrap flags=(unconfined) {
userns,
include if exists <local/bwrap>
}
EOF
```

The profile applies only to `bwrap` itself, not to the commands it runs inside the sandbox. Reload AppArmor to apply it:

```python
sudo systemctl reload apparmor
```

WSL1 does not support sandboxing because it lacks the required Linux namespace primitives. If you see `Sandboxing requires WSL2`, upgrade your distribution to WSL2 or run Claude Code without sandboxing. On WSL2, sandboxed commands cannot launch Windows binaries such as `cmd.exe`, `powershell.exe`, or anything under `/mnt/c/`. WSL hands these off to the Windows host over a Unix socket, which the sandbox blocks. If a command needs to invoke a Windows binary, add it to [`excludedCommands`](/docs/en/settings#sandbox-settings) so it runs outside the sandbox.

Enable sandboxing

You can enable sandboxing by running the `/sandbox` command:

```python
/sandbox
```

This opens a menu where you can choose between sandbox modes. If required dependencies are missing (such as `bubblewrap` or `socat` on Linux), the menu displays installation instructions for your platform. By default, if the sandbox cannot start (missing dependencies or unsupported platform), Claude Code shows a warning and runs commands without sandboxing. To make this a hard failure instead, set [`sandbox.failIfUnavailable`](/docs/en/settings#sandbox-settings) to `true`. This is intended for managed deployments that require sandboxing as a security gate.

Sandbox modes

Claude Code offers two sandbox modes: **Auto-allow mode**: Bash commands will attempt to run inside the sandbox and are automatically allowed without requiring permission. Commands that cannot be sandboxed (such as those needing network access to non-allowed hosts) fall back to the regular permission flow. Explicit deny rules are always respected, and `rm` or `rmdir` commands that target `/`, your home directory, or other critical system paths still trigger a permission prompt. Ask rules apply only to commands that fall back to the regular permission flow. **Regular permissions mode**: All bash commands go through the standard permission flow, even when sandboxed. This provides more control but requires more approvals. In both modes, the sandbox enforces the same filesystem and network restrictions. The difference is only in whether sandboxed commands are auto-approved or require explicit permission.

Auto-allow mode works independently of your permission mode setting. Even if you’re not in “accept edits” mode, sandboxed bash commands will run automatically when auto-allow is enabled. This means bash commands that modify files within the sandbox boundaries will execute without prompting, even when file edit tools would normally require approval.

Configure sandboxing

Customize sandbox behavior through your `settings.json` file. See [Settings](/docs/en/settings#sandbox-settings) for complete configuration reference.

Granting subprocess write access to specific paths

By default, sandboxed commands can only write to the current working directory. If subprocess commands like `kubectl`, `terraform`, or `npm` need to write outside the project directory, use `sandbox.filesystem.allowWrite` to grant access to specific paths:

```python
{
"sandbox": {
"enabled": true,
"filesystem": {
"allowWrite": ["~/.kube", "/tmp/build"]
}
}
}
```

These paths are enforced at the OS level, so all commands running inside the sandbox, including their child processes, respect them. This is the recommended approach when a tool needs write access to a specific location, rather than excluding the tool from the sandbox entirely with `excludedCommands`. When `allowWrite` (or `denyWrite`/`denyRead`/`allowRead`) is defined in multiple [settings scopes](/docs/en/settings#settings-precedence), the arrays are **merged**, meaning paths from every scope are combined, not replaced. For example, if managed settings allow writes to `/opt/company-tools` and a user adds `~/.kube` in their personal settings, both paths are included in the final sandbox configuration. This means users and projects can extend the list without duplicating or overriding paths set by higher-priority scopes. Path prefixes control how paths are resolved:

| Prefix | Meaning | Example |
|:------------------|:---------------------------------------------------------------------------------------|:--------------------------------------------------------------------------|
| `/` | Absolute path from filesystem root | `/tmp/build` stays `/tmp/build` |
| `~/` | Relative to home directory | `~/.kube` becomes `$HOME/.kube` |
| `./` or no prefix | Relative to the project root for project settings, or to `~/.claude` for user settings | `./output` in `.claude/settings.json` resolves to `<project-root>/output` |

The older `//path` prefix for absolute paths still works. If you previously used single-slash `/path` expecting project-relative resolution, switch to `./path`. This syntax differs from [Read and Edit permission rules](/docs/en/permissions#read-and-edit), which use `//path` for absolute and `/path` for project-relative. Sandbox filesystem paths use standard conventions: `/tmp/build` is an absolute path. You can also deny write or read access using `sandbox.filesystem.denyWrite` and `sandbox.filesystem.denyRead`. These are merged with any paths from `Edit(...)` and `Read(...)` permission rules. To re-allow reading specific paths within a denied region, use `sandbox.filesystem.allowRead`, which takes precedence over `denyRead`. When `allowManagedReadPathsOnly` is enabled in managed settings, only managed `allowRead` entries are respected; user, project, and local `allowRead` entries are ignored. `denyRead` still merges from all sources. For example, to block reading from the entire home directory while still allowing reads from the current project, add this to your project’s `.claude/settings.json`:

```python
{
"sandbox": {
"enabled": true,
"filesystem": {
"denyRead": ["~/"],
"allowRead": ["."]
}
}
}
```

The `.` in `allowRead` resolves to the project root because this configuration lives in project settings. If you placed the same configuration in `~/.claude/settings.json`, `.` would resolve to `~/.claude` instead, and project files would remain blocked by the `denyRead` rule.

Not all commands are compatible with sandboxing out of the box. Some notes that may help you make the most out of the sandbox:

- Many CLI tools require accessing certain hosts. As you use these tools, they will request permission to access certain hosts. Granting permission will allow them to access these hosts now and in the future, enabling them to safely execute inside the sandbox.
- `watchman` is incompatible with running in the sandbox. If you’re running `jest`, consider using `jest --no-watchman`
- `docker` is incompatible with running in the sandbox. Consider specifying `docker *` in `excludedCommands` to force it to run outside of the sandbox.

Claude Code includes an intentional escape hatch mechanism that allows commands to run outside the sandbox when necessary. When a command fails due to sandbox restrictions (such as network connectivity issues or incompatible tools), Claude is prompted to analyze the failure and may retry the command with the `dangerouslyDisableSandbox` parameter. Commands that use this parameter go through the normal Claude Code permissions flow requiring user permission to execute. This allows Claude Code to handle edge cases where certain tools or network operations cannot function within sandbox constraints.You can disable this escape hatch by setting `"allowUnsandboxedCommands": false` in your [sandbox settings](/docs/en/settings#sandbox-settings). When disabled, the `dangerouslyDisableSandbox` parameter is completely ignored and all commands must run sandboxed or be explicitly listed in `excludedCommands`.

Security benefits

Protection against prompt injection

Even if an attacker successfully manipulates Claude Code’s behavior through prompt injection, the sandbox ensures your system remains secure: **Filesystem protection:**

- Cannot modify critical config files such as `~/.bashrc`
- Cannot modify system-level files in `/bin/`
- Cannot read files that are denied in your [Claude permission settings](/docs/en/permissions#manage-permissions)

**Network protection:**

- Cannot exfiltrate data to attacker-controlled servers
- Cannot download malicious scripts from unauthorized domains
- Cannot make unexpected API calls to unapproved services
- Cannot contact any domains not explicitly allowed

**Monitoring and control:**

- All access attempts outside the sandbox are blocked at the OS level
- You receive immediate notifications when boundaries are tested
- You can choose to deny, allow once, or permanently update your configuration

Reduced attack surface

Sandboxing limits the potential damage from:

- **Malicious dependencies**: NPM packages or other dependencies with harmful code
- **Compromised scripts**: Build scripts or tools with security vulnerabilities
- **Social engineering**: Attacks that trick users into running dangerous commands
- **Prompt injection**: Attacks that trick Claude into running dangerous commands

Transparent operation

When Claude Code attempts to access network resources outside the sandbox:

1. The operation is blocked at the OS level
2. You receive an immediate notification
3. You can choose to:
- Deny the request
- Allow it once
- Update your sandbox configuration to permanently allow it

Security Limitations

- Network Sandboxing Limitations: The network filtering system operates by restricting the domains that processes are allowed to connect to. The built-in proxy does not terminate or perform TLS inspection on outbound traffic, so the contents of encrypted connections are not examined. You are responsible for ensuring that only trusted domains are allowed in your policy.

Allowing broad domains such as `github.com` can create paths for data exfiltration. Because the proxy makes its allow decision from the client-supplied hostname without inspecting TLS, code running inside the sandbox can potentially reach hosts outside the allowlist through traffic-shaping techniques. If your threat model requires stronger guarantees, configure a [custom proxy](#custom-proxy-configuration) that terminates TLS and inspects traffic, and install its CA certificate inside the sandbox. Stronger TLS-aware network isolation is an active area of development.

- Privilege Escalation via Unix Sockets: The `allowUnixSockets` configuration can inadvertently grant access to powerful system services that could lead to sandbox bypasses. For example, if it is used to allow access to a privileged host control socket this would effectively grant access to the host system. Users are encouraged to carefully consider any unix sockets that they allow through the sandbox.
- Filesystem Permission Escalation: Overly broad filesystem write permissions can enable privilege escalation attacks. Allowing writes to directories containing executables in `$PATH`, system configuration directories, or user shell configuration files can lead to code execution in different security contexts when other users or system processes access these files.
- Linux Sandbox Strength: The Linux implementation provides strong filesystem and network isolation but includes an `enableWeakerNestedSandbox` mode that enables it to work inside of Docker environments without privileged namespaces. This option considerably weakens security and should only be used in cases where additional isolation is otherwise enforced.

How sandboxing relates to permissions

Sandboxing and [permissions](/docs/en/permissions) are complementary security layers that work together:

- **Permissions** control which tools Claude Code can use and are evaluated before any tool runs. They apply to all tools: Bash, Read, Edit, WebFetch, MCP, and others.
- **Sandboxing** provides OS-level enforcement that restricts what Bash commands can access at the filesystem and network level. It applies only to Bash commands and their child processes.

Filesystem and network restrictions are configured through both sandbox settings and permission rules:

- Use `sandbox.filesystem.allowWrite` to grant subprocess write access to paths outside the working directory
- Use `sandbox.filesystem.denyWrite` and `sandbox.filesystem.denyRead` to block subprocess access to specific paths
- Use `sandbox.filesystem.allowRead` to re-allow reading specific paths within a `denyRead` region
- Use `Read` and `Edit` deny rules to block access to specific files or directories
- Use `WebFetch` allow/deny rules to control domain access
- Use sandbox `allowedDomains` to control which domains Bash commands can reach
- Use sandbox `deniedDomains` to block specific domains even when a broader `allowedDomains` wildcard would otherwise permit them

Paths from both `sandbox.filesystem` settings and permission rules are merged together into the final sandbox configuration. This [repository](https://github.com/anthropics/claude-code/tree/main/examples/settings) includes starter settings configurations for common deployment scenarios, including sandbox-specific examples. Use these as starting points and adjust them to fit your needs.

Advanced usage

Custom proxy configuration

For organizations requiring advanced network security, you can implement a custom proxy to:

- Decrypt and inspect HTTPS traffic
- Apply custom filtering rules
- Log all network requests
- Integrate with existing security infrastructure

```python
{
"sandbox": {
"network": {
"httpProxyPort": 8080,
"socksProxyPort": 8081
}
}
}
```

Integration with existing security tools

The sandboxed bash tool works alongside:

- **Permission rules**: Combine with [permission settings](/docs/en/permissions) for defense-in-depth
- **Development containers**: Use with [dev containers](/docs/en/devcontainer) for additional isolation
- **Enterprise policies**: Enforce sandbox configurations through [managed settings](/docs/en/settings#settings-precedence)

Best practices

1. **Start restrictive**: Begin with minimal permissions and expand as needed
2. **Monitor logs**: Review sandbox violation attempts to understand Claude Code’s needs
3. **Use environment-specific configs**: Different sandbox rules for development vs. production contexts
4. **Combine with permissions**: Use sandboxing alongside IAM policies for comprehensive security
5. **Test configurations**: Verify your sandbox settings don’t block legitimate workflows

Open source

The sandbox runtime is available as an open source npm package for use in your own agent projects. This enables the broader AI agent community to build safer, more secure autonomous systems. This can also be used to sandbox other programs you may wish to run. For example, to sandbox an MCP server you could run:

```python
npx @anthropic-ai/sandbox-runtime <command-to-sandbox>
```

For implementation details and source code, visit the [GitHub repository](https://github.com/anthropic-experimental/sandbox-runtime).

Limitations

- **Performance overhead**: Minimal, but some filesystem operations may be slightly slower
- **Compatibility**: Some tools that require specific system access patterns may need configuration adjustments, or may even need to be run outside of the sandbox
- **Platform support**: Supports macOS, Linux, and WSL2. WSL1 is not supported. Native Windows support is planned.

What sandboxing does not cover

The sandbox isolates Bash subprocesses. Other tools operate under different boundaries:

- **Built-in file tools**: Read, Edit, and Write use the permission system directly rather than running through the sandbox. See [permissions](/docs/en/permissions).
- **Computer use**: when Claude opens apps and controls your screen, it runs on your actual desktop rather than in an isolated environment. Per-app permission prompts gate each application. See [computer use in the CLI](/docs/en/computer-use) or [computer use in Desktop](/docs/en/desktop#let-claude-use-your-computer).