>_ LibrarianMCP

Turn any GitHub docs folder into a searchable MCP tool for your AI agents.

.mcp.json

 1{
 2  "mcpServers": {
 3    "librarian": {
 4      "type": "http",
 5      "url": "https://your-server.com/mcp/"
 6    }
 7  }
 8}

How it works

Point to docs

Give us a GitHub docs folder URL. We fetch all markdown files automatically.

We index & embed

Files are chunked by section, embedded with OpenAI, and stored in PostgreSQL with pgvector.

Agents search

Any MCP-compatible agent can semantically search the docs via a single HTTP endpoint.

Features

Semantic Search

Vector similarity search over docs using OpenAI embeddings and pgvector.

Multi-Project

Index multiple projects, search them independently. One server, all your docs.

Incremental Sync

Only re-embeds changed files on re-sync. Saves time and API costs.

Code Extraction

Code blocks are extracted as separate searchable chunks for better example discovery.

MCP Protocol

Standard Model Context Protocol. Works with Claude Code, Cursor, and any MCP client.

Relevance Scoring

Every result includes a relevance percentage so agents know how confident the match is.

<0ms

Search latency

Chunks indexed

Embedding dimensions

Get started

1. Run the serverbash

# Clone the repo
git clone https://github.com/your-org/librarian-mcp.git
cd librarian-mcp

# Configure environment
cp .env.example .env
# Edit .env with your OpenAI key

# Start everything
docker compose up -d

2. Connect your agentjson

{
  "mcpServers": {
    "librarian": {
      "type": "http",
      "url": "http://localhost:8000/mcp/"
    }
  }
}