cercano-local · git:20260410.2d55460 · 2026-04-10 · sha256 bcadecd94c32c7c0

cercano-local git:20260410.2d55460A

Immutable. This exact content never changes and is served at /api/v1/blob/bcadecd94c32c7c0.

---
name: cercano-local
description: Run prompts against local AI models via Cercano and Ollama. Use this for local inference — faster, private, and zero cost. Handles chat-style queries and agentic code generation with automatic validation. Offload summarization, explanation, code writing, and general LLM tasks to a local model instead of sending them to the cloud.
compatibility: Requires Cercano server running and connected to an Ollama instance.
---

# Cercano Local Inference

Run prompts against local AI models through Cercano's MCP interface. Cercano routes requests to Ollama for local inference.

## Important: Display the result

MCP tool results may not be visible to the user in the terminal. After calling the tool, you MUST output the full tool result text verbatim in your response so the user can see it.

## MCP Tool

**Tool name:** `cercano_local`

## Parameters

| Parameter | Type | Required | Description |
|-----------|------|----------|-------------|
| `prompt` | string | Yes | The prompt to run against local models. |
| `file_path` | string | No | Target file path for code changes. When provided with `work_dir`, enables the agentic code generation loop with validation. |
| `work_dir` | string | No | Working directory for code validation (go build/test). When provided with `file_path`, enables the agentic code generation loop. |
| `context` | string | No | Additional context such as existing code or file contents. |
| `conversation_id` | string | No | Conversation ID for multi-turn support across calls. |

## Modes

### Chat Mode
Provide only `prompt` (and optionally `context`) for a direct LLM call. The response is the model's text output.

### Agentic Code Generation Mode
Provide `prompt`, `file_path`, and `work_dir` to enable a generate-validate loop. Cercano will:
1. Generate code based on the prompt
2. Write it to the target file
3. Run validation (build/test) in the working directory
4. If validation fails, self-correct and retry

## Examples

**Chat query:**
```json
{
  "prompt": "What are the SOLID principles in software design?"
}
```

**Code generation with context:**
```json
{
  "prompt": "Add error handling to this function",
  "file_path": "internal/handler/auth.go",
  "work_dir": "/project",
  "context": "func Login(w http.ResponseWriter, r *http.Request) { ... }"
}
```

**Multi-turn conversation:**
```json
{
  "prompt": "Now refactor that to use the repository pattern",
  "conversation_id": "conv-abc123"
}
```