Skip to main content
Glama

NotebookLM MCP Server v1.0

Read this in other languages: Русский Production-ready, fully asynchronous Model Context Protocol (MCP) server for Google NotebookLM.

This server acts as a bridge between AI agents (like Claude Desktop, Cursor, Antigravity) and NotebookLM, allowing your AI to read your notebooks, interact with your sources, and generate reports.

Features

  • Asynchronous Execution: Fully utilizes asyncio for high concurrency handling.

  • Robust Execution: Wrap NotebookLM CLI commands with timeouts and proper error handling.

  • Thread Safety (Command Queue): Multiple simultaneous requests are queued safely via asyncio.Lock().

  • Advanced MCP Tools: Includes create_notebook, delete_notebook, add_source, list_sources, get_source_text, generate_audio, generate_report, and get_history.

  • Multi-language Support (i18n): All tool docstrings are in English, ensuring seamless compatibility with any LLM globally.

  • Network Mode (SSE Transport): Supports both Standard I/O (local) and SSE (network) transport protocols.

  • Retries & Caching: Temporary CLI failures are retried automatically. Read operations are cached.

  • Diagnostics: Built-in health_check() tool to verify CLI availability and authentication status.


Related MCP server: notebooklm-mcp

Installation

You can install this server either automatically (recommended) or manually.

This method automatically sets up a virtual environment, installs all dependencies (including notebooklm-py and Chromium for Playwright), and registers the server in your IDEs (Claude Desktop, Cursor, Antigravity).

  1. Clone or download this repository:

    git clone https://github.com/Vetal711/notebooklm-mcp.git
    cd notebooklm-mcp
  2. Run the auto-installer script:

    • Windows: Double-click install.bat (or run it in the terminal).

    • Mac/Linux: Run bash install.sh

  3. Authenticate with Google: Once the installation is complete, you must authenticate the CLI. Run the following command and follow the instructions in the browser window:

    • Windows: venv\Scripts\notebooklm.exe login

    • Mac/Linux: ./venv/bin/notebooklm login

  4. Restart your IDE/Agent (Claude Desktop, Cursor, or Antigravity) and the server will be available!


Method B: Manual Installation

If you prefer to configure everything manually or use a global Python environment:

  1. Clone the repository:

    git clone https://github.com/Vetal711/notebooklm-mcp.git
    cd notebooklm-mcp
  2. Create a virtual environment and install dependencies:

    python -m venv venv
    # On Windows:
    venv\Scripts\pip install -e .
    # On Mac/Linux:
    ./venv/bin/pip install -e .
  3. Install the Playwright browser driver:

    # On Windows:
    venv\Scripts\playwright install chromium
    # On Mac/Linux:
    ./venv/bin/playwright install chromium
  4. Authenticate:

    # On Windows:
    venv\Scripts\notebooklm.exe login
    # On Mac/Linux:
    ./venv/bin/notebooklm login
  5. Manual Configuration for MCP Clients: Instead of running the automated configuration script, you can manually add the following JSON block to your MCP client's configuration file (e.g., claude_desktop_config.json, cline_mcp_settings.json, or mcp_config.json).

    Make sure to replace /absolute/path/to/notebooklm-mcp with the actual absolute path to your cloned directory.

    {
      "mcpServers": {
        "NotebookLM": {
          "command": "/absolute/path/to/notebooklm-mcp/venv/bin/notebooklm-mcp",
          "args": []
        }
      }
    }

    (Note for Windows users: Use venv\\Scripts\\notebooklm-mcp.exe and escape backslashes in paths).

    Important: The server will automatically generate a .env file in the root folder on first run (or during installation) to handle absolute paths properly.


Troubleshooting

Error: Authentication expired If you receive this error in Claude/Cursor, it means the Playwright session's Google tokens have expired. WARNING: DO NOT run the login command inside the AI agent's chat or terminal! AI agents run in a background window station, which makes the popup browser window completely invisible, causing the login process to hang forever.

Solution:

  1. Open a standard, real OS terminal (or just double-click the script).

  2. Run login.bat (Windows) or login.sh (Mac/Linux) located in the project folder.

  3. A visible browser window will appear. Complete the login, and the window will close automatically.

  4. Ensure that your MCP client's configuration file (e.g., claude_desktop_config.json) includes the "cwd": "/absolute/path/to/notebooklm-mcp" parameter so it doesn't lose the session again.


Docker Deployment

To build and run via Docker:

docker build -t notebooklm-mcp .
docker run -i --rm -v ~/.notebooklm:/root/.notebooklm notebooklm-mcp

(Note: You need to mount the .notebooklm config directory to share your local authentication session).

Available Tools

12 tools
add_sourceA

Add a source (file, text, URL) to a notebook.

notebook_id: The ID of the notebook. content: The content (URL, absolute file path, or raw text). source_type: Optional. Type of source (url, text, file, youtube). Auto-detected if omitted.

Inputs: notebook_id (required), content (required), source_type. Returns: result

ask_notebookB

Ask a natural language question to a specific NotebookLM notebook.

notebook_id: The ID of the notebook. question: The question you want to ask based on the notebook's sources.

Inputs: notebook_id (required), question (required). Returns: result

create_notebookB

Create a new notebook in NotebookLM.

name: The name of the new notebook.

Inputs: name (required). Returns: result

delete_notebookA

Delete a notebook from NotebookLM.

notebook_id: The ID or name of the notebook to delete.

Inputs: notebook_id (required). Returns: result

generate_audioA

Generate an audio overview (podcast) from the notebook's sources. Note: This process can take several minutes to complete.

notebook_id: The ID of the notebook. prompt: Instructions for the audio overview (e.g., "focus on chapter 3").

Inputs: notebook_id (required), prompt (required). Returns: result

generate_reportC

Generate a written report (briefing doc, study guide, blog post) from the notebook. Note: This process can take several minutes to complete.

notebook_id: The ID of the notebook. prompt: Instructions for the report (e.g., "write a study guide for a 5th grader").

Inputs: notebook_id (required), prompt (required). Returns: result

get_historyC

Get the full conversation history (Q&A) for a specific notebook.

notebook_id: The ID of the notebook.

Inputs: notebook_id (required). Returns: result

get_source_textA

Get the full indexed text content of a specific source in a notebook.

notebook_id: The ID of the notebook. source_id: The ID of the source (use list_sources to find it).

Inputs: notebook_id (required), source_id (required). Returns: result

get_summaryC

Get the AI-generated summary of the notebook.

Inputs: notebook_id (required). Returns: result

health_checkA

Check the availability of the CLI and authentication status.

Inputs: no named parameters. Returns: result

list_notebooksA

Get a list of all NotebookLM notebooks.

Inputs: no named parameters. Returns: result

list_sourcesB

Get a list of all sources in a specific notebook.

Inputs: notebook_id (required). Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.3/5.0

    Scored across 12 tools

    Disambiguation5/5

    Each tool serves a distinct purpose: adding sources, asking questions, creating/deleting notebooks, generating audio/reports, retrieving history/summaries, and health checks. There is no overlap or ambiguity between tools.

    Naming Consistency4/5

    Most tools follow a verb_noun pattern (e.g., add_source, create_notebook), but 'health_check' is noun_verb, breaking the pattern. Otherwise, naming is consistent and clear.

    Tool Count5/5

    With 12 tools, the server covers the core NotebookLM workflows without being overwhelming. Each tool is necessary and well-scoped for the domain.

    Completeness3/5

    Missing operations include updating or deleting sources, updating notebook names, and a dedicated 'get_notebook' tool. Additionally, 'get_summary' lacks a notebook_id parameter, which is a notable gap.

    Related MCP Connectors

    Related MCP Servers