# Antigravity NotebookLM MCP
Advanced Model Context Protocol (MCP) server for Google NotebookLM.
Built for AI agents to autonomously research, manage knowledge bases, and generate multimedia artifacts using Google's NotebookLM engine.
## 🚀 Key Features
* **Robust Authentication**:
* **Browser Auth**: Safe, undetectable remote-debugging login flow (bypasses Google bot detection).
* **Manual Auth**: Fallback for headless environments.
* **Deep Research**: Autonomous web research agent integration with automated source importing.
* **Full Notebook Management**: CRUD operations for Notebooks.
* **Advanced Source Control**:
* Add Text, URLs, PDFs, and Drive files.
* **Sync**: Keep Drive sources up-to-date.
* **Check Freshness**: Monitor source status.
* **Studio & Artifacts**:
* Generate **Audio Overviews**, **Mind Maps**, **Quizzes**, **Study Guides**.
* Manage Studio: List and delete generated artifacts to keep projects clean.
* **Chat Configuration**: Set specific goals ("critique", "summary") and custom system prompts for your notebook.
## 🛠️ Installation
```bash
npm install
npm run build
```
## 🔐 Authentication (The Antigravity Way)
This server overcomes common bot detection issues using a dedicated Chrome automation workflow.
### Method 1: Browser Auth (Recommended)
This launches a specialized Chrome instance. You simply log in manually, and the MCP extracts the credentials automatically.
```bash
# Run the auth helper
node build/browser-auth.js
```
### Method 2: Manual Token
If you are on a headless server, you can manually inject cookies:
1. Open NotebookLM in your local browser.
2. Copy your `Cookie` header from DevTools (Network tab).
3. Use the `authenticate` tool in Claude/MCP with `method: "manual"`.
Credentials are securely stored in `~/.notebooklm-mcp/auth.json`.
## 📦 Usage with Claude Desktop
Add to your `claude_desktop_config.json`:
```json
{
"mcpServers": {
"notebooklm": {
"command": "node",
"args": ["/absolute/path/to/antigravity-notebooklm-mcp/build/index.js"]
}
}
}
```
## 🧰 Available Tools
| Tool | Action | Description |
|------|--------|-------------|
| **manage_notebook** | `list`, `get`, `create`, `rename`, `delete`, `configure_chat` | Full lifecycle management. `configure_chat` sets prompts/goals. |
| **manage_source** | `add`, `rename`, `delete`, `sync`, `check_freshness` | Manage knowledge sources. Supports syncing Drive files. |
| **manage_studio** | `list`, `delete` | Manage generated artifacts (Audio, Video, Mind Maps). |
| **query_notebook** | N/A | Ask questions grounded in your sources. |
| **perform_deep_research** | N/A | Execute multi-step deep web research and import findings. |
| **generate_artifact** | `audio`, `video`, `quiz`, `slides`, `mind_map`... | Generate multimedia content from your notes. |
| **authenticate** | `browser`, `manual` | Update session credentials. |
## 🏗️ Architecture
This project uses a direct reverse-engineered RPC client (`NotebookLMClient`) wrapped in an **MCP Server**.
It includes an **Orchestrator** layer for handling complex, multi-step asynchronous operations like Deep Research and Polling.
## ⚠️ Notes
* **Deep Research** operations can take 3-5 minutes. The server handles polling, but be patient.
* **Audio Generation** is a heavy task; ensure you have enough sources before generating.
## License
MIT
