Overleaf MCP

An MCP (Model Context Protocol) server that lets LLM agents (Claude Code, etc.) read and update Overleaf LaTeX papers via git.

Architecture

• overleaf_mcp.py — FastAPI server exposing read_paper and update_and_push_paper tools over HTTP
• main.py — stdio bridge that translates JSON-RPC (MCP protocol) to HTTP calls; this is what MCP clients connect to

Prerequisites

• Python 3.10+
• pdflatex (for LaTeX compilation)
• git

Setup

1. Clone your Overleaf project:

   git clone https://git.overleaf.com/<your-project-id> ~/path/to/your-paper

1. Configure the repository path:

   cp .env.example .env
   # Edit .env and set REPO_DIR to the absolute path of your local clone

1. Install dependencies:

   pip install -e .

Usage

Step 1: Start the MCP server

python overleaf_mcp.py

The server starts on http://localhost:8000 and exposes:

• GET /mcp/tools — list available tools
• POST /mcp/execute — execute a tool
• GET /health — health check

Step 2: Connect an MCP client

The bridge (main.py) speaks JSON-RPC over stdio, the standard MCP transport. Point your MCP client to it.

Claude Code (via `claude.json`):

{
  "mcpServers": {
    "overleaf": {
      "command": "python",
      "args": ["/path/to/overleaf_mcp/main.py"]
    }
  }
}

Or run the bridge directly for testing:

python main.py

Tools

Example

See mcp_sample.json for a sample request payload.