If you are an AI agent, LLM, or automated tool, a clean Markdown version of this page is available at https://www.kortex-notebooklm.com/use-cases/developers.md

Use your notebooks in your coding agent.

Kortex MCP lets Claude Code, Cursor, Codex, ChatGPT and other MCP clients search, read and add to your Gemini Notebook notebooks, with answers that cite their sources.

Get Kortex MCP Read the MCP docs

100,000+ users, rated 4.5 in 410 Chrome Web Store ratings.

Terminal

$claude mcp add --transport http kortex https://mcp.kortex-notebooklm.com/mcp

> Using my Payments API docs notebook, how do webhook retries work? Cite the sources.

kortex.ask_question "Payments API docs" answered

Failed deliveries are retried with exponential backoff for up to three days, and every retry carries the same event ID, so your handler should ignore IDs it has already processed.12

Example: Claude Code adds Kortex MCP, then asks a notebook of crawled API docs and gets a cited answer.

How do developers use Kortex?

Developers use Kortex MCP, so Claude Code, Cursor, Codex and other MCP clients can search, read and add to their Gemini Notebook (formerly NotebookLM) notebooks and get answers with citations. They fill those notebooks by crawling documentation sites and saving the AI chats where design decisions were made.

Why your agent can’t see your research

Your notes and docs are in Gemini Notebook. Your agent is in the terminal. Kortex MCP connects the two.

You want your agent to read your research, and Gemini Notebook has no public API for it.

Kortex MCP reaches your notebooks through the same signed-in path the extension uses in your browser, as 145+ tools any MCP client can call.
The docs you need are spread over a hundred pages.

Crawl a docs site from one address, to the depth you choose, into a notebook your agent can ask.
The reasoning behind a design lives in a Claude chat from March.

Send Claude or ChatGPT chats to a decisions notebook and search them with everything else.
An agent that can’t find the answer tends to make one up.

Asking a notebook through Kortex MCP answers from its sources and returns citations, so you can see where each point came from.
Handing an agent your Google login feels wrong.

Your Google login stays in your browser. Kortex never stores your password or cookies, and calls run through the extension on your own machine.

Set it up, then put it to work

Connect your assistant first, then save the chats worth keeping.

Connect Kortex MCP

One command in Claude Code, or one URL in any other client.

Install the Kortex extension in Chrome and sign in.
In Claude Code, run the command below. Your browser opens a consent page; sign in and choose what to allow.
For Cursor, Codex, ChatGPT, VS Code or Claude Desktop, give the client the same URL, or use an API key. See connect your client.
Keep Chrome running, minimised is fine. The first call after Chrome has been idle can take up to 30 seconds.

claude mcp add --transport http kortex https://mcp.kortex-notebooklm.com/mcp

Turn a docs site into a notebook

Cited answers from the docs you actually use, inside your editor.

In Bulk Import, start the web crawler at the docs’ first page and choose how many links deep to go.
Let Gemini Notebook finish processing the sources.
In your editor, ask: “Using my Payments API docs notebook, how do webhook retries work? Cite the sources.”
The answer comes back with citations to the pages it used.

Keep design decisions searchable

The why behind the code, searchable next to the docs.

When an architecture discussion in Claude or ChatGPT reaches a decision, click the Kortex button on the chat and send it to the project’s Decisions notebook.
Tag it with the service or the decision record’s number.
Later, press / in the Workspace to search every notebook, or ask your agent to search them over MCP.
If the discussion goes on, send the chat again and the source is updated in place.

Let the agent do the filing for you

Library chores done in a sentence, with write access you control.

Ask: “Create a notebook for the Billing v2 project, apply my project template, tag it #active and move it into Engineering.”
Or: “Compare my two research notebooks by source and move what only one has into the other.”
In Claude Code, the Kortex plugin adds ready-made recipes as slash commands.
Tools that change something are labelled as writes, and API key profiles decide what a client may call. See security and limits.

What reviewers say about these features

Quoted word for word and linked to where they were posted. Reviewers don’t say what they do for a living, so we picked reviews about the features on this page.

“Oh man, happen to know this extension from XDR. I should give 6 stars! Just one click and all chats imported to NBLM.”

Jeff Liou, on Chrome Web Store

“Easy to use and works well! Just click on the NotebookLM button in a chat, and you will be prompted to add the chat to a new project or existing one. The newly created source is formatted very cleanly, with prompts and responses indicated clearly (no extra random characters, etc.)”

Steve Lang, on Chrome Web Store

Docs and reading

MCP docs and comparisons

What developers ask

Something we missed? Ask us.

Does Kortex MCP need Chrome open?

Yes. Every call runs inside your Chrome extension, which holds your Google session. Chrome can be minimised, and no notebook tab needs to be open.

Which clients work with Kortex MCP?

Claude on the web and desktop through a custom connector, ChatGPT in developer mode, and coding tools including Claude Code, Cursor, VS Code, Windsurf, Cline, Codex CLI and Zed. Any client that supports remote HTTP MCP servers can connect; one that only runs local commands can wrap the URL with mcp-remote.

Does Kortex see my code or my Google password?

Kortex never stores your Google password or cookies. A tool call goes from your client to the Kortex endpoint and on to the extension on your machine, which talks to Gemini Notebook with your existing session. Your code only reaches a notebook if your agent adds it as a source.

Why does Chrome have to be open?

The extension holds your Gemini Notebook session, so with no extension running there is nothing to answer the call. Chrome can be minimised, and no notebook tab needs to be open.

How is it different from open-source NotebookLM MCP servers?

The best-known one, notebooklm-mcp-cli, is free, open source and signs in by saving your browser’s cookies. Kortex MCP runs each call through the Kortex extension in your own Chrome. The side-by-side comparison says where each one is the better pick.

Connect your agent to your notebooks.

Install the Kortex extension, then run one command to connect your assistant.