For years, the standard answer to "does NotebookLM have an API?" was no. Developers who wanted the product's outputs — the audio overviews, the source-grounded answers — automated it anyway, through community libraries that sign in with the user's own cookies and call endpoints Google never documented. Google now documents an official API for Gemini Notebook, the product's current name. It does considerably less than the unofficial tools, and that gap is the story.
Seven tools, all of them clerical
Google's documentation describes a REST interface at notebook.googleapis.com, mirrored by a remote MCP server at notebook.googleapis.com/mcp. The MCP page lists the complete tool set, and it is short: create a notebook, get one, list them, rename one, add a source, get a source, list sources. A notebook, in the API's own resource model, is a container: a server-assigned resource name, a display name and a creation time. Updating one means changing its display name, which is capped at 500 characters. That is the entire write surface.
What a source can be is equally narrow: a web URL, a YouTube video, plain text or Markdown. Calling get_source on one returns a processing state — PROCESSING, ACTIVE or FAILED — so an agent can poll until the source is ready to be used in the app. The API itself never uses it. The person who opens the notebook in the app does.
The missing half is the product
The API can retrieve notebook metadata and inspect sources, but it exposes no documented question-answering or generation method. There is no chat call, so the source-grounded question answering the product is known for is not available through it. There is no generation call of any kind: no audio overview, no video overview, no report, mind map, quiz, infographic or slide deck. The docs' overview page closes by inviting feedback and pointing to a roadmap of "upcoming capabilities". As documented today, the API covers the gathering, and stops before the outputs.
The one official route that generates anything is the separate Enterprise API in Google Cloud, and its limits are worth stating plainly. According to AutoContent API's comparison of the Enterprise documentation, it can create notebooks, add sources and make audio overviews — one per notebook at a time. Google Cloud lists Gemini Notebook for Enterprise at $9 per license per month, with a minimum of 15 licenses. It has no video, slides or infographics, and no method to download the audio it creates; the documented workflow directs users to open the notebook in a browser and download it from Studio by hand.
Preview terms, and who can actually use it
The consumer API is a public preview, and its account rules cut against the obvious audience. It requires a personal Google account: Workspace accounts are not yet supported, so the organizations most likely to build ingestion pipelines cannot use their work accounts. The user must have signed in to notebook.google.com at least once. Authentication is an OAuth 2.0 user token carrying a dedicated notebook scope, tied to a Google Cloud project — the API takes no API keys, so integrations must use the documented OAuth user-authorization flow rather than API-key authentication. A November 8 comparison by AutoContent API, compiled from Google's documentation, describes the preview as free during the preview period and puts the rate limit at 100 requests per minute per project and the notebook allowance at 100 to 500 per user, depending on the user's Google plan.
There was no launch event to read these terms against. No separate Google launch post was found: the documentation carries a November 2 update stamp, third-party comparisons had absorbed it by November 8, and daily briefings began listing an "official NotebookLM API" as news on November 11. Treat the limits above as preview documentation, compiled by a third party from Google's pages — the comparison notes that Google can change or stop a preview at any time, with no SLA.
What it is actually for
Stripped of the name, this is an ingestion pipe, and ingestion pipes have real uses. An agent that watches a set of feeds can file each item into a dated notebook without a human opening a browser tab. A research tool can assemble the source pack for a brief and hand a person a notebook that is ready to interrogate. The MCP server matters here: because the same seven tools are exposed over MCP, a coding agent can create and stock a notebook as one step inside a larger job, using the user's own account and permissions.
But the division of work should be clear before anyone builds on it. The machine files; the human asks. Every workflow built around the product's outputs — the podcast from a pile of PDFs, the quiz from a textbook — still happens inside the app, by hand, or through unofficial libraries running on undocumented endpoints. The comparison's flagship example is the open-source notebooklm-py library, which produces every Studio output and signs in with cookies from the user's own Google account; its own README labels it "Unofficial Library - Use at Your Own Risk". Developers comparing routes should price that honestly: the documented route is narrow by design, and everything beyond it runs on endpoints Google has not documented and can change at will.
Sources
- [1] Google for Developers, "Gemini Notebook API" (overview, last updated 2026-11-02 UTC)Read source
- [2] Google for Developers, "Notebooks | Gemini Notebook API" (resource model and REST methods)Read source
- [3] Google for Developers, "Model Context Protocol (MCP) | Gemini Notebook API" (remote MCP server, the 7 tools, personal-account and OAuth requirements)Read source
- [4] AutoContent API, "Does Gemini Notebook (NotebookLM) Have an API? Every Route, Compared" (November 8, 2026 short answer; preview price, rate and notebook limits compiled from Google's docs; Enterprise API terms from Google's docs of October 28 to November 5, 2026)Read source
- [5] Horizon daily briefing, November 11, 2026 (lists the NotebookLM API launch among the day's stories)Read source