NotebookLM MCP Server

This is an unofficial Model Context Protocol (MCP) server for Google NotebookLM, allowing AI agents and assistants (like Google Antigravity, Claude Code, Cursor, etc.) to query your Notebooks and retrieve citation-backed answers.

Prerequisites

• Python 3.10+
• A Google NotebookLM session cookie.

Installation

1. Clone this repository.
2. Initialize and activate a virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate
   

3. Install dependencies:

   pip install .
   

Configuration

You need to authenticate the unofficial API so it can access your Notebooks.

1. Authenticate via Playwright: Run the interactive login command provided by notebooklm-py:

   uv run notebooklm login
   # or if using a standard python venv:
   notebooklm login
   

   This will open a Chromium browser window where you can log in to your Google Account. Once logged in and on the NotebookLM page, close the browser. The session will be saved locally.

Usage

Start the MCP server over stdio using the command-line entry point:

uv run python -m mcp_notebooklm
# or if using standard python venv:
python -m mcp_notebooklm

Server Tools

This server exposes the following MCP tools:

• list_notebooks: Lists all your Notebooks (returns their IDs and Titles).
• get_notebook_sources: Retrieves the data sources for a specific notebook.
• ask_notebook: Passes a natural language query to a specific notebook and returns the AI-generated answer.

Using with Claude Desktop or Antigravity

Add this to your MCP settings configuration (mcp.json or equivalent):

{
  "mcpServers": {
    "notebooklm": {
      "command": "/path/to/your/virtualenv/bin/python",
      "args": [
        "-m",
        "mcp_notebooklm"
      ],
      "cwd": "/path/to/this/repo"
    }
  }
}