notebooklm-mcp-2026
MCP server for querying Google NotebookLM notebooks, enabling AI assistants to list notebooks, read sources, and ask questions about them.
README
<img src="notebooklm_mcp_hero.png" alt="NotebookLM MCP" width="100%">
notebooklm-mcp-2026
Secure MCP server for querying Google NotebookLM notebooks. Designed for use with Claude Code, Cursor, VS Code Copilot, and any MCP-compatible AI assistant.
Fork enhancements — This fork improves on julianoczkowski/notebooklm-mcp-2026 with multi-browser auth (including Helium), automatic session refresh, and a single
logincommand that verifies your account and configures MCP clients.
Watch on YouTube: https://youtu.be/xdI3uEA5rew?si=FkD0sdCZSFFWpjhy
<p align="left"> <a href="https://youtu.be/xdI3uEA5rew?si=FkD0sdCZSFFWpjhy"> <img src="https://img.youtube.com/vi/xdI3uEA5rew/maxresdefault.jpg" alt="Watch the video" width="100%"> </a> </p>
What it does
notebooklm-mcp-2026 gives AI assistants direct access to your Google NotebookLM notebooks. It runs as a local subprocess (stdio transport) — no HTTP server needed. Your AI assistant can list your notebooks, read source content, and ask the NotebookLM AI questions about your sources.
<p align="center"> <img src="flow_animation.gif" alt="NotebookLM MCP Flow" width="100%"> </p>
Quick Start
One command to log in, verify your account, and configure Cursor / Claude Code / VS Code. Works on macOS, Linux, and Windows.
Step 1: Install
From this fork (recommended):
git clone https://github.com/Faxtom/notebooklm-mcp-2026.git
cd notebooklm-mcp-2026
pip install -e .
Or with uv (upstream package):
curl -LsSf https://astral.sh/uv/install.sh | sh
uv tool install notebooklm-mcp-2026
Windows (PowerShell):
cd notebooklm-mcp-2026
python -m pip install -e .
Step 2: Log in (auth + verify + MCP setup)
notebooklm-mcp-2026 login --browser helium --method browser
This single command:
- Tries to import cookies from your browser (Helium, Chrome, Edge, Brave, Firefox, …)
- If no valid session is found, opens the browser so you can log in to notebooklm.google.com
- Verifies your account against the real NotebookLM API
- Auto-configures all detected MCP clients (Cursor, Claude Code, VS Code, …)
Using Helium? That's the recommended browser in this fork. Your normal Helium window can stay open — the interactive login uses an isolated profile.
Don't want auto-setup? Add
--skip-setupto only authenticate.
Step 3: Use it
Restart your MCP client and ask your AI assistant:
"List my NotebookLM notebooks"
That's it! You do not need to run setup separately after login.
Requirements
- A Chromium-based browser — Chrome, Edge, Brave, Helium, Chromium, Opera, or Vivaldi (for interactive login)
- Or Firefox / LibreWolf / Safari — cookie import only (no interactive login window)
- Python 3.11+
Don't have a browser?
- Helium (recommended): helium.computer · GitHub
- Chrome: google.com/chrome
- On Linux,
chromiumor the Helium.deb/ AppImage also work
Don't have Python?
If you used uv to install (recommended), you don't need to install Python separately — uv handles it for you.
If you prefer to install Python manually:
| Platform | Command |
|---|---|
| macOS | brew install python |
| Ubuntu / Debian | sudo apt install python3 |
| Arch / Manjaro | sudo pacman -S python |
| Fedora | sudo dnf install python3 |
| Windows | Download from python.org — tick "Add to PATH" during install |
Alternative Install Methods
Other install options:
# pipx (if you already have it)
pipx install notebooklm-mcp-2026
# pip (inside a virtual environment)
python -m venv .venv && source .venv/bin/activate
pip install notebooklm-mcp-2026
# From source
git clone https://github.com/Faxtom/notebooklm-mcp-2026.git
cd notebooklm-mcp-2026
pip install -e .
Authentication
notebooklm-mcp-2026 uses Google session cookies. No passwords are stored — only session cookies, a CSRF token, and a session ID.
Supported browsers
| Browser | Import cookies (--method browser) |
Interactive login (--method cdp) |
|---|---|---|
| Helium | ✅ | ✅ |
| Chrome | ✅ | ✅ |
| Edge | ✅ | ✅ |
| Brave | ✅ | ✅ |
| Chromium / Playwright | ✅ | ✅ |
| Opera / Vivaldi | ✅ | ✅ |
| Firefox / LibreWolf | ✅ | ❌ |
| Safari (macOS) | ✅ | ❌ |
Login methods
| Method | What it does |
|---|---|
auto (default) |
Try browser import → silent profile refresh → interactive CDP |
browser |
Import cookies; opens the browser automatically if no valid session |
cdp |
Open browser window for manual Google login |
profile |
Silent refresh from saved isolated browser profile |
import |
Load cookies from a JSON file |
Examples
# Recommended — Helium: import or open browser, then verify + configure MCP
notebooklm-mcp-2026 login --browser helium --method browser
# Auto-detect any installed browser
notebooklm-mcp-2026 login
# Interactive login with a specific browser
notebooklm-mcp-2026 login --browser edge --method cdp
# Import cookies from a JSON export
notebooklm-mcp-2026 login --method import --import-file cookies.json
# Login only, skip MCP configuration
notebooklm-mcp-2026 login --skip-setup
Session refresh
When cookies expire, the MCP server tries to refresh them automatically:
- Re-import from your installed browsers (Helium first)
- Re-read the isolated browser profile from a previous CDP login
- Only then ask you to run
loginagain
check_auth validates credentials with a real API call (not just a homepage check).
Helium paths (Windows)
| Item | Path |
|---|---|
| Executable | %LOCALAPPDATA%\imput\Helium\Application\chrome.exe |
| Profile | %LOCALAPPDATA%\imput\Helium\User Data |
Set a default browser with:
export NOTEBOOKLM_BROWSER=helium # macOS / Linux
$env:NOTEBOOKLM_BROWSER = "helium" # Windows PowerShell
Where credentials are stored
| Platform | Location |
|---|---|
| Linux | ~/.local/share/notebooklm-mcp-2026/auth.json |
| macOS | ~/Library/Application Support/notebooklm-mcp-2026/auth.json |
| Windows | %LOCALAPPDATA%\notebooklm-mcp-2026\auth.json |
Override with: NOTEBOOKLM_MCP_DATA_DIR=/custom/path
CLI Commands
| Command | Description |
|---|---|
notebooklm-mcp-2026 login |
Log in, verify account, and auto-configure MCP clients |
notebooklm-mcp-2026 setup |
Interactive setup wizard (optional — login does this automatically) |
notebooklm-mcp-2026 logout |
Remove stored credentials and browser profile |
notebooklm-mcp-2026 serve |
Start the MCP server over stdio (used by MCP clients) |
notebooklm-mcp-2026 status |
Show authentication and MCP client configuration status |
notebooklm-mcp-2026 doctor |
Diagnose common issues (browsers, auth, permissions) |
notebooklm-mcp-2026 version |
Print version |
Login flags
| Flag | Description |
|---|---|
--browser helium |
Use Helium (or chrome, edge, firefox, …) |
--method browser |
Import cookies; open browser if session is missing |
--method cdp |
Force interactive browser login |
--skip-setup |
Skip MCP auto-configuration after login |
--chrome-path PATH |
Path to browser executable |
--import-file PATH |
JSON cookie file (--method import) |
MCP Client Configuration
The setup command auto-configures your MCP client. You should not need to edit these files manually, but if you do:
<details> <summary>Claude Code — <code>~/.claude.json</code></summary>
{
"mcpServers": {
"notebooklm-mcp-2026": {
"command": "notebooklm-mcp-2026",
"args": ["serve"]
}
}
}
</details>
<details> <summary>Cursor — <code>~/.cursor/mcp.json</code></summary>
{
"mcpServers": {
"notebooklm-mcp-2026": {
"command": "notebooklm-mcp-2026",
"args": ["serve"]
}
}
}
</details>
<details> <summary>VS Code (Copilot) — <code>mcp.json</code></summary>
{
"servers": {
"notebooklm-mcp-2026": {
"command": "notebooklm-mcp-2026",
"args": ["serve"]
}
}
}
</details>
<details> <summary>Claude Desktop</summary>
Claude Desktop does not inherit your terminal's PATH, so you must use the full path to the executable.
First, find your executable path:
# macOS / Linux
which notebooklm-mcp-2026
# Windows (PowerShell)
where notebooklm-mcp-2026
Then edit your config file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
macOS example:
{
"mcpServers": {
"notebooklm-mcp-2026": {
"command": "/Users/YOUR_USER/.local/bin/notebooklm-mcp-2026",
"args": ["serve"]
}
}
}
Windows example:
{
"mcpServers": {
"notebooklm-mcp-2026": {
"command": "C:\\Users\\YOUR_USER\\.local\\bin\\notebooklm-mcp-2026.exe",
"args": ["serve"]
}
}
}
Replace YOUR_USER with your actual username, or paste the exact path from the which/where command above.
</details>
Available Tools (9)
| Tool | Description | Key Parameters |
|---|---|---|
login |
Refresh auth (browser import → CDP fallback) | timeout (default: 300s) |
check_auth |
Verify credentials via real API call | — |
list_notebooks |
List all notebooks with metadata | max_results (default: 50) |
get_notebook |
Get notebook details + source list | notebook_id |
list_sources |
List sources in a notebook | notebook_id |
get_source_content |
Get full text of a source | source_id |
query_notebook |
Ask the AI a question | notebook_id, query, source_ids?, conversation_id? |
add_source_url |
Add a URL/YouTube source | notebook_id, url |
add_source_text |
Add pasted text source | notebook_id, text, title? |
Typical workflow
1. list_notebooks → find the notebook ID you want
2. list_sources → see what sources are in it
3. query_notebook → ask questions about the sources
4. get_source_content → read raw source text if needed
Example output
When your AI assistant calls list_notebooks, it gets back structured data like this:
{
"status": "success",
"count": 2,
"notebooks": [
{
"id": "abc123-def456",
"title": "Research Notes",
"source_count": 3,
"is_owned": true,
"modified_at": "2026-01-15T10:30:00+00:00"
},
{
"id": "ghi789-jkl012",
"title": "Project Planning",
"source_count": 5,
"is_owned": true,
"modified_at": "2026-01-14T08:00:00+00:00"
}
]
}
And query_notebook returns:
{
"status": "success",
"answer": "Based on the sources, the main topics covered are...",
"conversation_id": "conv-uuid-123",
"turn_number": 1,
"is_follow_up": false
}
Follow-up conversations
query_notebook returns a conversation_id. Pass it back to ask follow-up questions in the same conversation context:
# First question
result = query_notebook(notebook_id="abc", query="What is the main topic?")
# result.conversation_id = "uuid-123"
# Follow-up
result = query_notebook(notebook_id="abc", query="Tell me more about that", conversation_id="uuid-123")
Troubleshooting
"Not authenticated" or "Cookies expired"
notebooklm-mcp-2026 login --browser helium --method browser
Browser import fails on Windows
Close the browser completely before import (the cookie database is locked while the browser runs). If import fails, --method browser will open the browser for interactive login automatically.
"Browser not found" error
Install Helium or Chrome, or specify the path:
notebooklm-mcp-2026 login --browser helium --method cdp
notebooklm-mcp-2026 login --chrome-path "C:\Users\YOU\AppData\Local\imput\Helium\Application\chrome.exe"
Run notebooklm-mcp-2026 doctor to see which browsers are detected.
Empty notebook list
Make sure you're logged into the correct Google account that has NotebookLM notebooks.
"Build label" errors
Google occasionally rotates their build label. Set the updated label:
NOTEBOOKLM_BL="boq_labs-tailwind-frontend_YYYYMMDD.XX_p0" notebooklm-mcp-2026 serve
Rate limit errors
NotebookLM free tier allows ~50 queries per day. Wait until the next day or upgrade.
Something else?
Run the diagnostic tool:
notebooklm-mcp-2026 doctor
Environment Variables
| Variable | Default | Description |
|---|---|---|
NOTEBOOKLM_MCP_DATA_DIR |
Platform default | Override data storage location |
NOTEBOOKLM_BROWSER |
— | Preferred browser (helium, chrome, firefox, …) |
NOTEBOOKLM_AUTH_REFRESH_COOLDOWN |
300 |
Seconds between silent session refresh attempts |
NOTEBOOKLM_BL |
boq_labs-tailwind-frontend_20260108.06_p0 |
Google build label |
NOTEBOOKLM_QUERY_TIMEOUT |
120.0 |
Query timeout in seconds |
Security
- No passwords stored — only Google session cookies
- File permissions — credentials saved with
0o600(owner read/write only) - Directory permissions — data directory created with
0o700(owner only) - No
eval/exec— no dynamic code execution anywhere - No
shell=True— browsers launched with explicit argument lists - Cookie filtering — only essential Google auth cookies are persisted
- Browser cleanup — browser process always terminated in
finallyblocks - Input validation — all tool parameters validated before use
- Timeouts — all HTTP requests have explicit timeouts
- CSRF protection — tokens passed in request body, auto-refreshed on expiry
Development
# Install dev dependencies
pip install -e ".[dev]"
# Run tests
pytest
# Lint
ruff check src/ tests/
Testing with MCP Inspector
The MCP Inspector lets you interactively test the server's tools in a web UI:
npx @modelcontextprotocol/inspector notebooklm-mcp-2026 serve
This opens a browser where you can call each of the 9 tools with custom parameters and inspect responses. You must run notebooklm-mcp-2026 login first.
Getting Help
- Questions? Open an Issue
- Found a bug? Open an Issue
- Want to contribute? See CONTRIBUTING.md
- Security issue? See SECURITY.md for responsible disclosure
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.