Skip to main content

Documentation MCP Server

LlamaFarm includes an MCP server that exposes the documentation to AI models, allowing them to search, read, and navigate the docs programmatically.

Why Use the MCP Server?​

  • AI-Assisted Development: Let Claude, GPT, or other AI assistants query LlamaFarm docs directly
  • IDE Integration: Use with Cursor, Claude Code, or other MCP-enabled tools
  • Accurate Answers: AI gets real-time access to current documentation
  • Context Retrieval: Models can search for relevant configuration examples

Quick Setup​

1. Build the Server​

cd docs/website/mcp
npm install
npm run build

2. Configure Your AI Tool​

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "llamafarm-docs": {
      "command": "node",
      "args": ["/path/to/llamafarm/docs/website/mcp/dist/index.js"]
    }
  }
}

Claude Code (.mcp.json in project root):

{
  "mcpServers": {
    "llamafarm-docs": {
      "command": "node",
      "args": ["docs/website/mcp/dist/index.js"]
    }
  }
}

Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "llamafarm-docs": {
      "type": "stdio",
      "command": "node",
      "args": ["docs/website/mcp/dist/index.js"]
    }
  }
}

Available Tools​

list_docs​

List all available documentation files:

Tool: list_docs
Arguments: { "category": "rag" }  // Optional filter

Returns file paths, titles, and descriptions.

read_doc​

Read a specific documentation file:

Tool: read_doc
Arguments: { "path": "rag/parsers.md" }

Returns the full markdown content.

search_docs​

Search across all documentation:

Tool: search_docs
Arguments: {
  "query": "PDFParser configuration",
  "max_results": 5
}

Returns matching files with line numbers and context.

get_toc​

Get the documentation structure:

Tool: get_toc
Arguments: {}

Returns a hierarchical table of contents.

Example Interactions​

Once configured, you can ask your AI assistant questions like:

"Search the LlamaFarm docs for how to configure CrossEncoderRerankedStrategy"

The AI will use the search_docs tool to find relevant documentation, then read_doc to get the full content.

"List all the RAG documentation pages"

The AI will call list_docs with category: "rag".

"Show me the parsers reference documentation"

The AI will call read_doc with path: "rag/parsers.md".

Using with LlamaFarm​

You can also configure LlamaFarm itself to use the docs MCP server:

mcp:
  servers:
    - name: llamafarm-docs
      transport: stdio
      command: node
      args:
        - docs/website/mcp/dist/index.js

runtime:
  models:
    - name: assistant
      provider: ollama
      model: llama3.1:8b
      mcp_servers: [llamafarm-docs]

This allows LlamaFarm's chat agents to query the documentation when answering questions.

Development​

To modify the MCP server:

cd docs/website/mcp

# Development mode with hot reload
npm run dev

# Inspect with MCP Inspector
npm run inspect

# Build for production
npm run build

How It Works​

  1. The server scans docs/website/docs/ for all markdown files
  2. Extracts titles and descriptions from frontmatter
  3. Provides full-text search with relevance scoring
  4. Exposes docs as both MCP tools and resources

Source Code​

See docs/website/mcp/ for the full implementation.