🌌 Gemini Notebook MCP

The Enterprise-Grade Model Context Protocol (MCP) Server for Google NotebookLM.
Bridge the gap between Google's state-of-the-art Gemini grounding engine and your autonomous AI agent squad (Antigravity 2.0, Claude Code, Cursor, Windsurf, OpenCode, Jules).

Features • Installation • Client Config • Tool Reference • Architecture • SDLC Ground-Truth

🚀 Overview

Gemini Notebook MCP (v2.1.0) is the next-generation, actively maintained MCP server connecting AI assistants directly to Google NotebookLM.

While traditional AI workflows consume hundreds of thousands of tokens injecting huge raw PDFs, transcripts, and codebases into context windows, Gemini Notebook MCP unlocks a 90% Token Reduction through NotebookLM's native Map-Reduce retrieval. It provides autonomous AI agents with instant access to synthesized domain knowledge, full conversation histories, notes dumps, and direct local file exports for Studio artifacts (generated React/TSX components, Python scripts, study guides, reports, and audio podcasts).

Related MCP server: NotebookLM MCP

✨ Key Features

🎨 1. Studio Panel Artifact Extraction & Auto-Downloads

• list_studio_items: Inspect all studio items generated by NotebookLM (Mind Maps, Study Guides, Briefing Docs, FAQs, Timelines, Audio Overviews, Custom Code Artifacts).

• get_studio_item: Read the full content and metadata of any studio artifact on demand.

• download_studio_item: Trigger native file downloads for any studio asset directly into your local workspace directory.

• export_studio_artifacts: Bulk export all studio assets (code .tsx, .ts, .py, .json, .md, images, audio) in one unified command.

💬 2. Complete Conversation & Thought-Chain Reconstruction

• get_chat_history: Retrieve complete structured chat turns between user and NotebookLM.

• export_chat_history: Dump entire notebook conversation transcripts to disk in structured JSON or Markdown formats.

• Deep Grounding: Extracts citations, referenced page fragments, and underlying model reasoning threads.

📝 3. Comprehensive Notes Knowledge Base

• list_notes: List all saved notes with tags and source anchors.

• get_note: Read individual note contents by ID or title.

• export_notes: Export all notes (e.g. 100+ research summaries) into markdown files for offline analysis and agent indexing.

🔍 4. Multimodal Sources & Media Retrieval

• list_sources: Retrieve all uploaded PDFs, Google Docs, YouTube transcripts, and web links.

• download_source_images: Extract and download source figures, architectural diagrams, screenshots, and visual assets directly to disk.

• add_source: Ingest new URLs, text snippets, or files programmatically into any notebook.

🎙️ 5. Audio Overview (Deep Dive Podcasts)

• generate_audio: Trigger generation of dual-host conversational podcast overviews.

• get_audio_status: Real-time progress monitoring (pending, generating, ready).

• download_audio: Download generated .wav / .m4a audio files directly to your storage.

🛡️ 6. Anti-Detection Stealth Engine (Patchright)

• Built on Patchright (hardened Playwright fork) with stealth flags, persistent human fingerprints, and zero browser automation leakage (navigator.webdriver patched).

• Headless execution with auto-fallback to Chromium channel.

🏗️ Architecture

flowchart TD
    subgraph Agents["Autonomous AI Agents & IDEs"]
        AG[Antigravity 2.0 Orchestrator]
        CL[Claude Desktop / Claude Code]
        CR[Cursor / Roo Code / Windsurf]
        OC[OpenCode Frontend & Jules Cloud]
    end

    subgraph MCP["Gemini Notebook MCP Server (v2.1.0)"]
        Router["JSON-RPC Tool Router (stdio / Streamable-HTTP)"]
        AuthMgr["Auth & Session Manager (State & Fingerprint)"]
        StudioExt["Studio Artifact & File Downloader Engine"]
        ChatHist["Chat & Thought-Chain Extractor"]
        NotesMgr["Notes & Ground-Truth Exporter"]
        PatchEng["Patchright Stealth Browser Core"]
    end

    subgraph Google["Google Cloud Infrastructure"]
        NLM["Google NotebookLM"]
        GeminiCore["Gemini 2.5 Pro / Flash Grounding Core"]
        Storage["NotebookLM Studio & Media Storage"]
    end

    Agents <-->|MCP Protocol| Router
    Router --> AuthMgr
    Router --> StudioExt
    Router --> ChatHist
    Router --> NotesMgr
    StudioExt & ChatHist & NotesMgr --> PatchEng
    PatchEng <-->|Secure Persistent Session| NLM
    NLM <--> GeminiCore
    NLM <--> Storage

📦 Installation

Option 1: Run via NPX (Recommended)

npx gemini-notebook-mcp@latest

Option 2: Clone & Build from Source

git clone https://github.com/vn002187444/gemini-notebook-mcp.git
cd gemini-notebook-mcp
npm install
npm run build

🔐 One-Click Authentication Setup

To authenticate with your Google Account once:

# In the repository root
npm run login

A dedicated browser window will launch. Log in to your Google Account and visit NotebookLM. The session state and credentials are automatically saved to your OS user data directory:

• Windows: %LOCALAPPDATA%\notebooklm-mcp\Data\browser_state\state.json

• macOS: ~/Library/Application Support/notebooklm-mcp/browser_state/state.json

• Linux: ~/.local/share/notebooklm-mcp/browser_state/state.json

Subsequent runs run completely headless and require no user interaction.

⚙️ Client Configuration

1. Antigravity 2.0 & AGY CLI

Add to your mcp_servers configuration in ~/.gemini/antigravity/mcp.json or project configuration:

{
  "mcpServers": {
    "notebooklm": {
      "command": "node",
      "args": ["d:/NotebookLM Knowledge/notebooklm-mcp/dist/index.js"],
      "env": {
        "BROWSER_CHANNEL": "chrome"
      }
    }
  }
}

2. Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "notebooklm": {
      "command": "npx",
      "args": ["-y", "gemini-notebook-mcp@latest"]
    }
  }
}

3. Cursor & Windsurf

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "notebooklm": {
      "command": "node",
      "args": ["<PATH_TO_REPO>/dist/index.js"]
    }
  }
}

🛠️ MCP Tools Reference

1. Studio Artifacts & Downloads (v2.1.0 New)

2. Chat Conversation History & Thought Extraction (v2.1.0 New)

3. Notes Knowledge Base (v2.1.0 New)

4. Notebook & Source Management

5. Grounded Q&A & Audio Overviews

🎯 SDLC Ground-Truth Protocol

When working in an agentic pair-programming workflow (e.g. Antigravity 2.0 SDLC), use the NotebookLM Ground-Truth & Map-Reduce Protocol:

1. Avoid Raw Dumps: Never load 50,000 lines of documentation or legacy code directly into your prompt. Ingest it as a NotebookLM source.

2. 4-Pass Deep Inquiry: Use ask_question to query core logic, architecture schemas, and edge cases.

3. Studio Asset Pull: Use export_studio_artifacts to pull all code snippets (.tsx, .py, .ts) and assets synthesized by NotebookLM directly into the local repo (docs/nlm_ground_truth_code/).

4. Zero Hallucination Verification: Cross-reference generated code against the DOM-level citations returned by get_chat_history.

👨‍💻 Development & Testing

# Watch mode for rapid TypeScript development
npm run dev

# Run full TypeScript check and build
npm run check

# Code formatting & linting
npm run format
npm run lint

📄 License & Acknowledgements

• Licensed under the MIT License.

• Built for the autonomous AI development era by vn002187444.

• Special thanks to the original notebooklm-mcp prototype foundations by Gérôme Dexheimer.