notebooklm-mcp

The project builds on Gérôme Dexheimer’s NotebookLM MCP server (Please Prompto!, MIT licence), to which I added a REST API, Studio generation and account rotation, for a doctoral literature review run in batches of over 1,000 questions. Its main limitation: it goes through the NotebookLM web app’s internal calls, which Google does not document and can change without notice; the project is not affiliated with Google, and NotebookLM’s quotas apply.

No playable demo: it needs a Google account and a NotebookLM notebook.

A service with no API, made callable

NotebookLM answers from the sources placed in a notebook, and each answer points to the passages it rests on. Google offers no public API for it. The project therefore takes two routes: the batchexecute calls the web app itself uses, and an automated browser (Playwright) as a fallback when those calls change.

A citation makes checking possible: you go and read the passage. It does not guarantee the answer is right; the answer still has to be read critically.

Since July 2026 Google has called the product “Gemini Notebook”. It is the same service; the project keeps its name.

Getting started

In Claude Code, through the roomi-fields/claude-plugins marketplace (Node 18 or later):

/plugin marketplace add roomi-fields/claude-plugins
/plugin install notebooklm@roomi-fields

Then, once, sign in to Google from a terminal; a Chrome window opens:

npx -y -p @roomi-fields/notebooklm-mcp notebooklm-mcp-setup-auth

For the REST API (n8n, Zapier, Make, curl):

git clone https://github.com/roomi-fields/notebooklm-mcp.git
cd notebooklm-mcp
npm install && npm run build
npm run setup-auth
npm run start:http
curl -X POST http://localhost:3000/ask \
  -H 'Content-Type: application/json' \
  -d '{"question": "Summarize chapter 3", "notebook_id": "your-id", "source_format": "json"}'

A Docker image is also available, for a server or a NAS.

What it does

• Ask a question and get the answer with its citations, in five formats: none, inline, footnotes, JSON, expanded with the quoted excerpt.
• Generate Studio content: audio overview, video, infographic, report, presentation, data table, flashcards, quiz, mind map; in any of 81 accepted languages.
• Download the audio (WAV), the video (MP4), the infographic (PNG); delete generated content (content_delete).
• Manage sources: add a PDF, a DOCX, a URL, a YouTube video, a Drive document; read the exact text NotebookLM indexed (source_read).
• Run batches: several accounts in rotation when a daily quota runs out, automatic re-authentication, resuming an interrupted batch.
• Keep the answers: vault_batch writes them to disk, as Markdown and as JSON following a published schema, readable afterwards by rtfm.

The REST API has 33 endpoints, described in OpenAPI 3.1.

Common issues

• Signing in from the assistant fails: some clients, Claude Desktop among them, cut off a tool call that runs too long. Sign in from a terminal (#27).
• Account: use a Google account dedicated to automation; Google may detect automated use.
• Google Workspace accounts: recognised since 2.2.1.
• Content generated in the wrong language: the interface language overrode the language argument up to 3.1.x; fixed.

Where it stands

Version 3.2.0 of 4 September 2026, 60 releases since November 2025; continuous integration passes. 184 stars and 56 forks as of 28 September 2026; 2,913 npm downloads from 29 August to 27 September 2026, by npm’s own counter, which includes automated installs. Editing notes is not done yet. What is new goes in the repository’s changelog.

Repository · Documentation · [[NotebookLM]] · [[RTFM]]

Thanks to Gérôme Dexheimer (@PleasePrompto) for the original project, and to Khizar Jamshaid Iqbal, Kazik Pietka, Rui Ruiberriz, @he0xwhale, @eminsnow and Julien Cantoni for their fixes and additions.