Gemini Notebook (formerly NotebookLM) API: Key and Pricing
Is there a NotebookLM API? Not for personal accounts. See the Enterprise API at $9 a license a month, Gemini API File Search and MCP servers for your agents.
You want a script, an agent or an app to talk to your notebooks. So you search for a Gemini Notebook (formerly NotebookLM) API and find a pile of half-answers. Here’s the straight one: personal accounts get no public API. That said, it isn’t the end of the story. Google documents an enterprise API and a deprecated podcast API, and the Gemini API offers a build-it-yourself route. MCP servers cover much of the rest.
Key Takeaways
- Google documents an API only for Gemini Notebook Enterprise, which costs $9 per license per month after a trial (Google Cloud, 2026).
- The standalone Podcast API is deprecated.
- The Gemini API’s File Search tool can power a cited alternative.
- MCP servers connect AI agents without an official API.
Does Gemini Notebook have a public API?
No, not for personal accounts. Google’s only documented Gemini Notebook API belongs to Gemini Notebook Enterprise on Google Cloud, and its setup steps require Enterprise licenses before you make a single call (Google Cloud, 2026). Free, Plus, Pro and Ultra accounts have nothing official to call. As a result, every consumer integration you’ll find works around the app, not through it. In fact, a public request for an official, OAuth-based NotebookLM MCP server is open on Google’s MCP repository (GitHub, 2026). Its author argued that an official server would bring the “stability and security required for professional workflows”. Why does that matter to you? Workarounds can break whenever Google changes the web app, so anything you build on one needs a plan for the day it stops.
You’re not alone in wanting one. We analyzed 1,451 Google autocomplete suggestions for these topics in October 2026, and typing “notebooklm api” alone brought up “notebooklm api key”, “notebooklm api pricing”, “notebooklm api free” and “notebooklm api for claude”. Similarly, 32 of the 776 people who answered our Kortex MCP waitlist survey (Sep-Oct 2026) asked for a real API in their own words. One wanted to “programmatically add sources to NLM into client notebooks”.
If what you really want is Claude or ChatGPT working with your notebooks, you don’t need an official API.
Let your AI assistant use your notebooks
Kortex MCP gives Claude, ChatGPT, Cursor and other MCP clients 145+ notebook tools, each call run by the Kortex extension in your own Chrome. It's included with Kortex Pro and Lifetime.
What can the Enterprise API do, and what does it cost?
The Gemini Notebook Enterprise API is Google’s v1alpha API for notebooks in the Google Cloud edition. It manages notebooks, sources and audio, but it doesn’t chat. Google’s docs list methods to create, get, list recently viewed, delete and share notebooks, add or upload sources, and create or delete audio overviews (Google Cloud, 2026). Specifically, every call goes to a v1alpha endpoint under Google’s Pre-GA terms, which “might have limited support”. In addition, calls carry an OAuth access token from gcloud auth print-access-token. That’s why there’s no “NotebookLM API key” to copy from a settings page.
| Area | Documented methods |
|---|---|
| Notebooks | notebooks.create, get, listRecentlyViewed, batchDelete, share |
| Sources | notebooks.sources.batchCreate, uploadFile, get, batchDelete |
| Audio overviews | notebooks.audioOverviews.create, delete |
Sources can be Google Docs, Slides, raw text, web pages or YouTube videos, plus uploads such as PDF, DOCX or MP3 files. However, a notebook can hold only one audio overview at a time (Google Cloud, 2026). What’s missing matters more. No documented method sends a question and returns a cited answer. In other words, the API automates setup, while people still read and chat in the web app.
Google’s three API pages document 11 methods, and none of them sends a question to a notebook. In particular, the role for sharing carries the old name: Cloud NotebookLM User (Google Cloud, 2026).
Access costs money and admin work. Google’s product page lists Gemini Notebook for Enterprise at $9 per license per month after a 30-day trial, with a 15-license minimum, so the smallest paid setup runs $135 a month (Google Cloud, 2026). Its licensing docs describe a 14-day trial of 5,000 licenses instead, so confirm the trial terms with Google before you budget. On top of licenses, someone has to set up Gemini Notebook Enterprise in a Cloud project and handle OAuth tokens. That’s routine for a Cloud team. For one person who wants an agent inside their own notebooks, though, it’s a lot of setup for an API that can’t ask a question. Our Gemini Notebook for business guide covers when Enterprise is worth it.
Is there a Gemini Notebook podcast API?
There was, and it’s deprecated. Google Cloud’s standalone Podcast API turned source material into MP3 episodes without needing a notebook or a license. Its docs say: “The Podcast API is deprecated. Google isn’t allowlisting new customers” (Google Cloud, 2026). In addition, input had to stay under 100,000 tokens, and the API offered two lengths: SHORT, typically 4-5 minutes, and STANDARD, around 10 minutes. Want NotebookLM-style hosts without that API? Enterprise customers can call the audio overview method on a notebook instead. Alternatively, anyone can script a two-voice episode with the Gemini API’s text-to-speech, which handles up to 2 speakers in one request (Google AI for Developers, 2026). For example, a script can pass a written dialogue for two prebuilt voices and save the audio that comes back.
Can you build your own Gemini Notebook with the Gemini API?
Partly. File Search is a Gemini API tool that imports, chunks and indexes your files, then grounds answers in the passages it retrieves. Answers can cite the document parts and page numbers they used, files can be up to 100 MB each, and storage is free (Google AI for Developers, 2026). You pay for embeddings at indexing time and for retrieved text as ordinary input tokens, and the free tier holds 1 GB in total. Google calls this Retrieval Augmented Generation, the approach Patrick Lewis and colleagues described in 2020. In fact, they found it produced “more specific, diverse and factual language” than a model answering only from what it had memorized (arXiv, 2020). Thus File Search gives you the cited-answer half of Gemini Notebook, as long as you write the app around it.
| Gemini Notebook | Gemini API + File Search | |
|---|---|---|
| How you use it | Web and mobile app | Your own code, with an API key |
| Citations | Inline, in the chat | Annotations with document parts and page numbers |
| Audio and video sources | Supported | Not supported |
| Studio outputs | Audio, video, slides, quizzes and more | You build them yourself |
| Cost | Free tier, then Google AI plans | Indexing embeddings plus model tokens |
| Size cap | Source limits per notebook, by plan | 100 MB per file, 1 GB total on the free tier |
The trade is control for polish. File Search gives you the retrieval half of Gemini Notebook with an API attached, but none of the Studio. Therefore, pick it when your app needs cited answers from your documents, not audio or slides. That said, plan around one limit: it can’t be combined with Grounding with Google Search in the same request (Google AI for Developers, 2026).
How do MCP servers connect agents to Gemini Notebook?
Through third-party servers, since Google doesn’t offer one of its own. The Model Context Protocol (MCP) is “an open-source standard for connecting AI applications to external systems” (MCP, 2026), and every Gemini Notebook MCP server is a third-party project built without an API. The community ones take two approaches. PleasePrompto’s notebooklm-mcp, with 3,400+ GitHub stars, drives a real Chrome browser with Patchright. However, it’s archived, and its README warns that it “may stop working when the upstream services change” (GitHub, 2026). The gemini-notebook-mcp-cli project (formerly notebooklm-mcp-cli) calls Google’s internal APIs instead. Its README calls them undocumented and liable to change, and it needs cookies extracted from your browser (GitHub, 2026). Notably, both depend on Google’s web app staying as it is.
In both cases your Google session stays on your machine. Consequently, the agent can act as you. Is that a trade you’re happy with? Kortex MCP takes a third route, covered below. Our Gemini Notebook MCP guide compares the options in depth.
How do you get your data out without an API?
Often, “API” really means “get my stuff out”. Gemini Notebook exports a note to Google Docs: hover over the note in the Sources panel, select More, then Export to Docs (Gemini Notebook Help, 2026). Edits you make there don’t sync back to the notebook, and sharing permissions don’t carry over to the new file either. Meanwhile, sources and chats are harder. Google documents a bulk export for Workspace admins, whose Data Export tool covers chat history, uploaded content, generated notes and audio overviews (Google Workspace Admin Help, 2026). In contrast, personal accounts get no per-notebook export of sources or chats at all. Our guide to exporting Gemini Notebook covers every method.
Want notes in Notion or Obsidian rather than Docs? Our guide to moving notebook notes into Notion or Obsidian shows an assistant doing it through Kortex MCP, included with Kortex Pro and Lifetime. For files on your own disk, though, an export button beats any API.
Export sources, notes and chats in one click
Kortex adds export buttons to Gemini Notebook: every source in a notebook as PDF, Markdown, text or a ZIP, and chats as Markdown, text, PDF or JSON. Free on Chrome.
Which route fits your project?
Pick by who owns the data and what you need back. The Enterprise API suits companies already buying Enterprise licenses (15 minimum), File Search suits developers building an app, and MCP suits one person who wants an assistant inside existing notebooks (Google Cloud, 2026).
| You want to… | Best route |
|---|---|
| Create notebooks and load sources for a whole company | Gemini Notebook Enterprise API |
| Ask cited questions over your documents from your own app | Gemini API with File Search |
| Generate a two-voice audio episode in code | Gemini API text-to-speech |
| Let Claude or ChatGPT use your existing notebooks | An MCP server |
| Save sources, notes and chats as files | Native export plus an extension |
Notice the gap in that table. No official route lets a personal account’s notebooks answer questions from code. That’s the obvious job for an API, and it’s the one Google leaves empty. Hence the interest in MCP servers, which reach the notebooks you already have.
How does Kortex MCP connect your agent to Gemini Notebook?
Kortex MCP is a hosted MCP server that gives Claude, ChatGPT, Cursor, VS Code and other MCP clients 145+ notebook tools (Kortex MCP docs, 2026). We built it as a bridge: your client calls a Kortex endpoint, and the Kortex extension runs each call in your own Chrome, with the Google session you’re already signed in to. It comes with Kortex Pro and Lifetime. There’s no stealth browser and no exported cookie file. Instead, Chrome has to be running with the extension signed in, and Kortex never stores your Google password or cookies (Kortex MCP docs, 2026). As a result, the first call after a quiet spell can take up to 30 seconds while the extension wakes up.
You need three things:
- A Kortex Pro or Lifetime account.
- Chrome open, with the Kortex extension signed in.
- An MCP client, such as Claude Code, Claude Desktop, ChatGPT on the web or Cursor.
Connecting takes one URL: add https://mcp.kortex-notebooklm.com/mcp as a Streamable HTTP server, then sign in on the Kortex consent page it opens. Scripts and CI can send a Kortex API key as Authorization: Bearer kx_mcp_... instead. Notably, new keys default to read-only (Kortex MCP docs, 2026).
The tools fill the gap above: create notebooks, add sources, ask questions with citations, tag, compare and generate. Our guide to getting cited notebook answers in Claude or ChatGPT shows the question-and-answer loop end to end. Additionally, the tools can run on a clock. For instance, your assistant can save a job that asks a notebook, saves the answer or makes an Audio Overview every hour, day or week (Kortex workflow docs, 2026). The client setup guide covers Claude Code, Claude Desktop, ChatGPT, Cursor, Windsurf, Zed and more.
Kortex has a free plan, and Pro costs $9.99 a month, or $4.92 a month billed yearly (Kortex pricing, 2026). Don’t need an agent? Kortex’s import button in ChatGPT, Claude, Gemini and Perplexity sends a chat to a notebook, and chats export as Markdown or JSON.
There’s no consumer API, and you don’t need one. Kortex brings import, export and automation to Gemini Notebook, and Kortex MCP, included with Pro and Lifetime, puts your notebooks within reach of your AI agent.
Give your agent your notebooks, no API needed
Kortex MCP lets Claude, ChatGPT and Cursor create notebooks, add sources and ask cited questions through the Kortex extension in your own Chrome. It's included with Kortex Pro and Lifetime, and Kortex itself is free on Chrome.
Frequently asked questions
Does Gemini Notebook have an API?
Not for personal accounts. Google documents an API only for Gemini Notebook Enterprise on Google Cloud, which needs Enterprise licenses. Free, Plus, Pro and Ultra users on notebook.google.com have no official way to call their notebooks from code.
Can I get a NotebookLM API key?
Not for a personal account, because the consumer app has no public API. The Gemini Notebook Enterprise API signs calls with a Google Cloud OAuth access token from gcloud, not a simple key, and it needs Enterprise licenses. A Gemini API key works with File Search, not with your notebooks.
What can the Gemini Notebook Enterprise API do?
It can create, get, list recently viewed, delete and share notebooks, add or upload sources such as Google Docs, web pages and YouTube videos, and create or delete an audio overview. It’s a v1alpha preview, and none of its 11 documented methods sends a question to a notebook.
How much does the NotebookLM API cost?
The only official one, the Gemini Notebook Enterprise API, needs Gemini Notebook for Enterprise licenses, which cost $9 per license per month after a 30-day trial, with a 15-license minimum. The Gemini API’s File Search charges for indexing and model tokens instead, and its storage is free.
Is there a NotebookLM podcast API?
There was. Google Cloud’s standalone Podcast API turned under 100,000 tokens of context into an MP3 episode, but it’s deprecated and Google isn’t allowlisting new customers. Alternatives are the Enterprise audio overview method or the Gemini API’s text-to-speech, which handles 2 speakers per request.
Can I build my own NotebookLM with the Gemini API?
Partly. The Gemini API’s File Search tool indexes your files and answers with citations and page numbers. Storage is free, and you pay for indexing and tokens. Files can be up to 100 MB each, with 1 GB in total on the free tier. You’d build audio and study outputs yourself.
How do I connect Claude or ChatGPT to Gemini Notebook?
Through an MCP server. Community servers automate a browser or call undocumented internal APIs with your cookies, and one of the best-known is archived on GitHub. Kortex MCP, a Kortex Pro and Lifetime feature, gives MCP clients 145+ notebook tools that run through the Kortex extension in your own Chrome.
Editorial note: fact-checked against the official help pages and docs linked above on October 26, 2026. Spotted something out of date? Contact us and we will fix it, or read more about Kortex.
