KastorFlowUpdated October 30, 2026
NotebookLM MCP: connect your notebooks to an AI agent
A NotebookLM MCP (Model Context Protocol) server lets an AI app such as Claude Desktop, Claude Code or Cursor find your notebooks, read their sources and add new ones. Open-source servers sign in with copies of your Google cookies. KastorFlow's MCP server is free and open source. It works through the extension on your own computer, only inside the notebooks you choose.
Gemini Notebook is NotebookLM's new name. Everything here applies to both.
Your options
| Option | How it signs in | Watch out for |
|---|---|---|
| Open-source NotebookLM MCP servers (GitHub) | Copies of your Google cookies, taken from a Chrome window it opens or pasted in by you | Unofficial: can break when Google changes its pages. Whoever holds the cookies can act as you. You sign in again when they expire. |
| Google's NotebookLM Enterprise API | Google Cloud credentials | For organizations on NotebookLM Enterprise, not personal accounts. |
| KastorFlow MCP | Doesn't copy anything: it works through KastorFlow in your own Chrome | Chrome must be open. Free on GitHub, with Free-plan tools. Pro and Pro+ tools need those plans. |
How many servers there are and how often they're downloaded: NotebookLM API and MCP statistics
Set up KastorFlow's free MCP server
- Install the bridge once. You need Node.js 20 or later. Run
npx kastorflow-notebooklm-mcp setup. - In KastorFlow, open Settings → AI agents and choose Connect an agent. Pick the notebooks and what the agent may do. KastorFlow shows the setup for your AI app.
- Add it to your AI app, such as Claude Desktop, Claude Code or Cursor, using the setup KastorFlow shows.
Keep Chrome open with KastorFlow while the agent works. Full setup, including Windows: KastorFlow NotebookLM MCP on GitHub.
How KastorFlow's MCP works
Your agent app → kastorflow-mcp → KastorFlow in Chrome → Gemini Notebook
- Everything runs on your computer. Nothing listens on a network port and nothing passes through KastorFlow's servers.
- You choose which notebooks each agent may use and what it may do. Pause or revoke an agent in Settings → AI agents and it stops at once.
- An agent can't delete, merge, move or refresh anything. Sharing a notebook, sending anything to a connected tool or to Google Docs, and publishing a podcast episode wait until you approve them in KastorFlow.
- The agent never receives your Google sign-in, your KastorFlow account or the credentials of your connected tools.
- Text that comes back from your sources is marked as your content, never as instructions to the agent.
What an agent can do through KastorFlow
| Kind | For example |
|---|---|
| Read | List notebooks and sources, read a source, ask a notebook (Pro) |
| Collect | Create a notebook, add sources, import a YouTube playlist, start research |
| Organize | Rename, tag, write notes, copy sources |
| Generate | Studio artifacts, a briefing, a notebook audit |
| Asks you first | Share a notebook, send to a connected tool, publish a podcast episode |
| Keep a record | Save research and propose decisions you approve, then pick them up from another agent |
The free server on GitHub has the 28 Free-plan tools. The full companion, kastorflow-mcp, has 51, and each one needs the KastorFlow plan its feature needs. An agent sees only the tools you allow.
Use NotebookLM in Claude and Claude Code
- Through an MCP server: Claude Desktop and Claude Code connect to MCP servers, so any option in the table above works. KastorFlow's MCP server is free and open source on GitHub (Apache-2.0) and runs on your computer through the KastorFlow extension in your own Chrome, so Claude never gets your Google cookies. Add it to Claude Desktop or Claude Code like any MCP server, then choose in KastorFlow which notebooks Claude may use. Among the open-source servers, notebooklm-mcp-cli is a CLI, an MCP server and agent skills in one; it signs in with your browser's cookies and, in its own words, uses internal APIs that may not be aligned with Google's terms.
- Through a Claude skill: a skill is a folder of instructions and scripts that Claude loads when it needs them. The best-known NotebookLM skill, notebooklm-skill, drives a Chrome window to ask your notebooks from Claude Code. It was archived in September 2026, as was the same author's notebooklm-mcp.
- Without either: save the answer as a note in Gemini Notebook and paste it into Claude, or save the Claude chat back to a notebook. Save Claude chats to NotebookLM
From each repository's README and Anthropic's What are skills?, read October 25, 2026.
KastorFlow's MCP server is free and open source. Connect your first agent in a few minutes.
Get it on GitHubQuestions
Is there an official NotebookLM MCP server?
We know of none from Google for personal accounts as of October 2026. Google's API is for NotebookLM Enterprise on Google Cloud. For a personal account, the options are open-source servers or an extension such as KastorFlow.
Is it safe to give an MCP server my Google cookies?
Your cookies let whoever holds them act as your Google account until they expire. If you use a cookie-based server, read its code and keep it on your own computer. KastorFlow's MCP doesn't copy them: the agent never receives your Google sign-in.
Which AI apps work with KastorFlow's MCP?
Claude Desktop, Claude Code, Cursor and other apps that support MCP.
Can I use NotebookLM in Claude?
Yes, through an MCP server or a Claude skill: Claude Desktop and Claude Code can then list your notebooks, ask them and add sources. Google has no official connector for personal accounts, so these are open-source tools or an extension such as KastorFlow. NotebookLM in Claude
Does Chrome need to be open?
Yes. KastorFlow's MCP works through the extension, so Chrome must be open with KastorFlow. While it's closed, the agent is told KastorFlow isn't connected.
What does KastorFlow's MCP cost?
The MCP server on GitHub is free and works on every KastorFlow plan, with the Free-plan tools. Pro and Pro+ tools need those plans and the full companion, kastorflow-mcp. Pricing