local-docs-mcp

Local documentation search for AI assistants, from Git repos and local directories

What It Does

An MCP server that gives AI assistants access to locally indexed documentation. It indexes GitHub repositories (via sparse checkout) and local directories into a SQLite database with FTS5 full-text search. Token budgeting keeps results within AI context window limits.

Quick Start

Add to your MCP client:

Claude Code (~/.claude.json), Cursor (Settings > MCP), Cline (VS Code settings):

{
  "mcpServers": {
    "local-docs": {
      "command": "npx",
      "args": ["-y", "local-docs-mcp@latest", "stdio"]
    }
  }
}

VS Code Copilot (.vscode/mcp.json):

{
  "servers": {
    "local-docs": {
      "command": "npx",
      "args": ["-y", "local-docs-mcp@latest", "stdio"]
    }
  }
}

Example

The AI assistant discovers and indexes documentation on the fly. For example:

1. You ask: "How does Entra Connect hybrid identity sync work?"
2. AI calls list_repos - no relevant docs found
3. AI researches the GitHub repo and calls add_docs with:
   • url: "https://github.com/MicrosoftDocs/entra-docs"
   • paths: ["docs/identity/hybrid/"]
   • alias: "entra-hybrid"
4. AI calls list_repos to check indexing progress
5. AI calls search_docs with "hybrid identity sync" and answers with the indexed docs

No config files needed. The AI handles discovery, indexing, and search.

MCP Tools

| Tool | Description | |------|-------------| | search_docs | Full-text search across all indexed docs (or a specific repo) | | list_repos | List available documentation sources, status, and indexing progress | | browse_docs | Explore doc tree structure and headings | | update_docs | Re-index to pull latest changes | | add_docs | Add a new git repo or local directory as a doc source | | remove_docs | Remove a documentation source |

How It Works

• Sparse checkout - only fetches specified paths from GitHub repos, minimizing bandwidth
• FTS5 full-text search - SQLite FTS5 with BM25 ranking for relevance
• Markdown chunking - splits docs by heading, preserving document structure
• Token budgeting - fits results within AI context window limits
• Background indexing - add_docs returns immediately while indexing runs async
• Auto-refresh - stale repos are re-indexed automatically on server startup

More

• CLI reference and standalone install
• Architecture and design

AI Agent Integration

A skills/local-docs-mcp/SKILL.md is included for AI agent platforms that support the agentskills.io specification. Copy or symlink the skills/local-docs-mcp/ directory to your skills directory for enhanced tool usage guidance.

Contributing

PRs welcome. Run go test ./... before submitting.