# Kortex for AI agents and assistants > How AI agents read this site (Markdown twins, llms.txt), connect Kortex MCP to Gemini Notebook (formerly NotebookLM), and the rules for crawling and training. https://www.kortex-notebooklm.com/agents # Kortex for AI agents and assistants People send assistants to read this site, and Kortex MCP lets those assistants work in Gemini Notebook (formerly NotebookLM). This page is the map for both. [Connect Kortex MCP](/docs/mcp-quickstart) [Read llms.txt](/llms.txt) ## Read this site Every page has a Markdown twin: the same words as the HTML, built from it at each deploy, without the layout. 1. ### Add .md to the path `/features` becomes `/features.md`, and the home page is `/index.md`. 2. ### Ask for Markdown Send `Accept: text/markdown` and the page’s own URL answers in Markdown. 3. ### Follow the link Every HTML page names its twin in a `Link` header and an alternate `` tag. ```bash curl -H "Accept: text/markdown" https://www.kortex-notebooklm.com/features ``` ### Files written for agents [`/llms.txt`](/llms.txt) The index. Every page with a one-line summary, grouped by topic and a checkout link for each plan. Start here. [`/llms-full.txt`](/llms-full.txt) The product pages and every docs page in one file, for when you can only make one fetch. [`/index.md`](/index.md) The home page as Markdown. Every other page works the same way: add .md to its path. [`/sitemap-index.xml`](/sitemap-index.xml) Every URL on the site, docs included. [`/.well-known/mcp/server-card.json`](/.well-known/mcp/server-card.json) Kortex MCP described for clients: its URL, transport, sign-in and tools, rebuilt from the live server at each deploy. [`/.well-known/api-catalog`](/.well-known/api-catalog) An API catalog (RFC 9727) pointing to the MCP server, its card, its docs and the status page. [`/robots.txt`](/robots.txt) Who may crawl, and what they may do with what they read. ## Connect Kortex MCP Kortex MCP is a server that gives an assistant tools to work with someone’s Gemini Notebook library. It works in Claude, ChatGPT, Cursor, VS Code and any client that speaks MCP over HTTP. Chrome has to be running with the Kortex extension signed in: the extension holds the Google session, so calls run there and no Google password or cookie leaves the browser. Server URL. Sign in with OAuth, or send a Kortex API key. ``` https://mcp.kortex-notebooklm.com/mcp ``` In Claude Code, one command: ```bash claude mcp add --transport http kortex https://mcp.kortex-notebooklm.com/mcp ``` - [Quickstart (opens in a new tab)](/docs/mcp-quickstart) - [Connect your client (opens in a new tab)](/docs/mcp-clients) - [Create an API key (opens in a new tab)](/docs/mcp-api-key) - [Tools reference (opens in a new tab)](/docs/mcp-tools) - [Security and limits (opens in a new tab)](/docs/mcp-security) ## What it can do With Kortex MCP connected, an assistant works in the notebooks directly. Kortex covers the notebooks. The assistant’s other connectors cover Gmail, Slack, Drive or files. - **Ask across a library.** Answers based on your sources, with citations, from one notebook or from many at once. - **Fill notebooks from the chat.** URLs, YouTube videos, RSS feeds, Google Docs, pasted text and PDFs become sources without opening a tab. - **Organize in bulk.** Tag, file, rename, merge, dedupe and clean up hundreds of notebooks and sources in one instruction. - **Generate and export.** Start Audio Overviews, study guides, video overviews and data tables, then export them to Markdown, Docs or Sheets. - **Run on a schedule.** Save a workflow once and it runs on a schedule, such as a Monday digest of last week’s sources. Things people ask for, from the [MCP docs](/docs/mcp-server): - “Go through my inbox for every email from my landlord this year and put them in one notebook.” - “Read all 200 of my sources and tag the ones that are actually about pricing.” - “Every Monday, summarise last week’s sources and save the digest as a new source.” ## The rules Short, and the same for every crawler, agent and person. Crawl all of it [`robots.txt`](/robots.txt) has one rule for everyone: `User-agent: *`, `Allow: /`. No crawler is singled out. Index, answer, train Its `Content-Signal` line reads `search=yes`, `ai-input=yes`, `ai-train=yes`: you may list these pages in search, use them to answer questions, and train models on them. One version of every page We never change a page because of who asks for it. Markdown is a format, not a different story: the twin carries the same words as the HTML. What we count The server notes which AI agent fetched which page, so we know which pages agents rely on. The response is the same either way. Ask before you act Kortex MCP labels its tools as read-only, changing or deleting. Confirm deletions with the person, and buy, sign up or share only when they asked you to. Buying for someone Once someone has chosen a plan, [`llms.txt`](/llms.txt) has a checkout link for each plan. The plan goes to the Kortex account with the email they pay with. ## Facts to get right What people most often need from an answer about Kortex. What it is An independent Chrome extension for Gemini Notebook. Kortex is not made by Google. The name Gemini Notebook was called NotebookLM until July 16, 2026. Both names mean the same Google product. Price Current plans and prices are on the [pricing section](/#pricing). Reach 100,000+ users, rated 4.5 from 410 Chrome Web Store ratings, as of November 5, 2026. Where it imports from ChatGPT, Claude, Gemini, Perplexity, X, Reddit, LinkedIn, YouTube, Google Docs, Grok, Kimi, Genspark, OpenEvidence, plus any web page. Contact [kortexnotebooklm@gmail.com](mailto:kortexnotebooklm@gmail.com)