# SandBase Harness

> Local-first, self-hosted runtime for durable and auditable AI agent execution.

- Project: https://github.com/sandbaseai/sandbase-harness
- Documentation: https://github.com/sandbaseai/sandbase-harness/blob/main/docs/README.md
- Installation: https://github.com/sandbaseai/sandbase-harness/blob/main/llms-install.md
- Showcase: https://github.com/sandbaseai/sandbase-harness/blob/main/docs/showcase.md
- Official MCP Registry: https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.sandbaseai%2Fsandbase-harness
- Latest tagged release: v0.3.8
- Official DeepSeek Harness Showcase: https://github.com/deepseek-ai/deepseek-harness/discussions/1918
- Verified dshbase listing: https://dshbase.com/plugins/sandbase-harness/
- Independent self-hosting guide (SSD Nodes): https://www.ssdnodes.com/learn/self-host-sandbase-agent-runtime
- Agent Runtime landscape: https://github.com/sandbaseai/awesome-agent-runtime
- Verified community references: dshbase, Awesome Agents, abordage/awesome-mcp, and merged entries in awesome-agent-runtime, Awesome Agent Cortex, and Awesome AI Devtools.

## Discovery

- MCP Market: https://mcpmarket.com/server/sandbase-harness
- deepseek-plugin.org: https://deepseek-plugin.org/plugins/sandbaseai/sandbase-harness
- DeepseekPlugin: https://deepseekplugin.org/en/plugins/sandbaseai-sandbase-harness
- DSH Plugin Directory: https://dshplugin.app/plugins/sandbase-harness
- DSH Plugin Hub: https://dshpluginhub.dev/en/plugins/sandbaseai/sandbase-harness
- DSH Directory: https://dsh.directory/plugins/sandbaseai/sandbase-harness
- DSH Harness: https://dsharness.io/en/plugins?search=sandbase-harness
- DSH Plugin: https://dshplugin.me/?q=sandbase-harness
- DSH Plugin: https://dsh-plugin.org/plugins/sandbaseai/sandbase-harness
- Awesome repository index: https://awesome.lvtd.dev/repos/?topic=dsh-plugin
- Awesome DSH Plugin: https://github.com/awesome-dsh-plugin/awesome-dsh-plugin
- Awesome DeepSeek Harness: https://github.com/0xsline/awesome-deepseek-harness
- Awesome DeepSeek Harness — ecosystem list: https://github.com/fendouai/awesome-deepseek-harness
- DSHarness 101 Plugin Radar: https://dsharness101.com/plugins/
- DeepSeekDocs Ecosystem: https://deepseekdocs.com/en/ecosystem
- Awesome Agents: https://github.com/kyrolabs/awesome-agents
- Sifted Awesome AI Agents — Agent Runtime Top 100: https://github.com/sifted-network/sifted-awesome-ai-agents/blob/main/top100/Agent%20Runtime.md
- Arnon-hs Open Source — MCP projects: https://github.com/Arnon-hs/open-source/blob/main/mcp/README.md
- SandBase Awesome Agent Runtime: https://github.com/sandbaseai/awesome-agent-runtime
- abordage/awesome-mcp: https://github.com/abordage/awesome-mcp
- cccakeee/awesome-dsh-plugins: https://github.com/cccakeee/awesome-dsh-plugins
- anbeime/skill — Skills index: https://github.com/anbeime/skill
- Awesome DeepSeek Harness Plugins: https://github.com/Zhiyuan-Fan/Awesome-DeepSeek-Harness-Plugins
- Hermes Ecosystem — SandBase stack: https://github.com/ksimback/hermes-ecosystem/blob/main/projects/sandbaseai/cli.html
- AgentStack: https://www.agentstack.live/mcp/io.github.sandbaseai/sandbase-harness
- MCP Servers Live: https://linny006.github.io/mcp-servers-live/r/sandbaseai/sandbase-harness/
- DSH X-Ray: https://unstone.github.io/dsh-xray/p/sandbaseai__sandbase-harness.html
- DSH Plugins: https://github.com/HackSing/dsh-plugins
- Awesome DSH Plugins 2026: https://github.com/Herdeny/awesome-dsh-plugins-2026
- MCP Repository: https://mcprepository.com/sandbaseai/sandbase-harness
- OpenAgentSkill (code-review): https://www.openagentskill.com/skills/sandbaseai-sandbase-harness-code-review
- PluginBench: https://pluginbench.com/mcp/io.github.sandbaseai/sandbase-harness
- DSH Hub: https://dshhub.dev/plugins/sandbase-harness
- DSH Packs: https://www.dshpacks.com/plugins/sandbaseai-sandbase-harness/
- dshbase: https://dshbase.com/plugins/sandbase-harness/
- FindHarness: https://findharness.com/plugins/sandbaseai-sandbase-harness
- DSH Plugin Directory: https://dsh-plugin.github.io/directory.html
- DSH Plugin Registry: https://github.com/dshplugin-app/deepseek-harness-plugins
- dsh-market: https://dshmarket.com/p/sandbaseai/sandbase-harness/
- dshplugin.dev: https://dshplugin.dev/plugins/sandbaseai-sandbase-harness

These independent directory pages are discovery mirrors; this repository and its
release metadata remain the source of truth.

## Capabilities

- Persistent agent sessions, event streams, artifacts, memory, and resumable work
- Local, Docker, Kubernetes, and self-hosted worker sandbox providers
- MCP toolsets and a six-tool stdio MCP bridge
- Managed credentials, permission policies, approvals, audit logs, and replay
- Built-in local Console and `/v1` HTTP API
- OpenAI, Anthropic, MiniMax, and OpenAI-compatible model providers, including DeepSeek V4
- SQLite and file-backed local-first storage with no required hosted control plane

## MCP bridge

- Docker image: `ghcr.io/sandbaseai/sandbase-harness-mcp:0.3.8`
- Container package: https://github.com/orgs/sandbaseai/packages/container/package/sandbase-harness-mcp
- Tools: `list_agents`, `create_session`, `run_session`, `get_session`, `list_artifacts`, `stop_session`
- Runtime endpoint: `MANAGED_AGENTS_URL` (default `http://127.0.0.1:3000`)
- Optional authentication: `MANAGED_AGENTS_API_KEY`

## Limits and safety

- The MCP image is a bridge to a running Harness API; it is not a standalone model provider.
- The local sandbox shares the host OS user boundary; use Docker or Kubernetes for stronger isolation.
- Configure model-provider credentials through the runtime's environment or secret-management path; do not commit keys.

## Repository guides

- Usage: https://github.com/sandbaseai/sandbase-harness/blob/main/docs/usage.md
- Deployment: https://github.com/sandbaseai/sandbase-harness/blob/main/docs/deployment.md
- MCP source: https://github.com/sandbaseai/sandbase-harness/blob/main/src/mcp/server.ts