Google ships Gemini Notebook API in public preview: REST and MCP server, source ingestion only

Google has published a public-preview (pre-GA) REST API and an MCP server for Gemini Notebook, the product formerly called NotebookLM, for use with personal Google accounts. According to Japanese-language outlet @IT (Deep Insider Brief), which tested it hands-on, the API automates adding sources such as web pages and YouTube videos, but cannot yet ask questions of a notebook or generate outputs.

The source reports that Gemini Notebook was renamed from NotebookLM in July 2026. Before this release, individuals had no official API. Programmatic access required either the paid Enterprise API or browser-automation tools. Google's documentation was updated between October 29 and November 2, and @IT infers the release timing from those dates. Google has also published a roadmap with planned features.

What the API can do

• Notebooks: create, list, get and rename.
• Sources: add four types: web page URL, YouTube video URL, pasted text and Markdown text.
• Ingestion status: ingestion is asynchronous. A source moves through PROCESSING, ACTIVE or FAILED, and the client polls until it is ready.
• Authentication: OAuth 2.0 with the scope https://www.googleapis.com/auth/notebook. A standard gcloud login token does not include this scope, so calls fail without it.
• Samples: Python, JavaScript and curl examples are in the documentation.

Shortened YouTube URLs (youtu.be) are normalized by the API to the www.youtube.com/watch?v= form.

MCP server

The MCP endpoint is https://notebook.googleapis.com/mcp, using Streamable HTTP (JSON-RPC 2.0). It exposes seven tools: create_notebook, get_notebook, list_notebooks, update_notebook, create_source, get_source and list_sources.

Requests carry two headers: an Authorization: Bearer OAuth token and x-goog-user-project with a Google Cloud project ID. Tokens expire after one hour and must be reissued. The documentation includes a prompt to hand to a coding agent, which automates project selection, API enablement, scoped authentication and MCP server registration.

Hands-on results

The @IT writer passed that prompt to Codex and added three YouTube videos to a notebook through MCP. Each reached ACTIVE in 4 to 6 seconds. A web page took about 9 seconds and Markdown text about 6 seconds. These timings are client-side observations that include tool-call intervals and network time. The sources displayed in the browser with titles, thumbnails and links, as if added manually.

Not yet supported

• Querying a notebook and retrieving answers
• Generating outputs such as video or audio overviews
• Deleting notebooks or sources (returns HTTP 501 Not Implemented)
• Uploading binary files such as PDFs
• Google Drive or Google Docs URLs as sources (returns an error)
• Google Workspace accounts (returns HTTP 403 Forbidden)

Google lists these as planned on its roadmap, which @IT describes as short, about six items, with no order or timing. Google states the roadmap is informational and may change.

Prerequisites and caveats

Users must sign in once to notebook.google.com beforehand, or calls return HTTP 404. A Google Cloud project and the gcloud CLI are required, and notebook and per-source word limits apply. As a pre-GA release, support is limited and incompatible changes are possible. This is separate from the paid Gemini Notebook Enterprise API.

What this means

For individual users, source ingestion can now be scripted or delegated to an MCP-capable agent without browser automation. @IT's author argues that fixed API calls are easier to constrain than an agent driving a browser UI, so the scope of delegated operations is narrower and safer. The missing query and generation endpoints mean end-to-end pipelines still need manual steps in the browser. @IT's author says he hopes the API coverage widens.