Documentation built for AI agents as well as people
A hosted MCP server on your own docs domain, auto-generated llms.txt and llms-full.txt, and content structured for clean retrieval, so ChatGPT, Claude, Cursor, and your customers all read the same current source.
The world's best product teams trust Documentation.AI
From next-gen start-ups to established enterprises.
Your product has a second audience now
A growing share of the questions about your product are answered by a model, from whatever it can find.
- Models answer from stale training data
- Your API changed in March. The assistant your customer is pairing with learned it a year before that and states it with total confidence.
- Scraped HTML loses the structure
- Navigation, sidebars, and interactive widgets arrive as noise, so the chunk that reaches the model is half boilerplate.
- No feed to point anything at
- There is nowhere to send an AI client that wants your current documentation, so it crawls and guesses instead.
- You cannot see any of it
- AI agent traffic looks like bot traffic, so nobody can say whether models are reading your docs or inventing them.
Three ways in, all pointing at the same current source
llms.txt for discovery, clean Markdown for retrieval, and an MCP server for agents that would rather search than crawl.
- llms.txt and per-page Markdown are generated for you, and can be switched on or off in Site Config
- A Reader MCP server on your own docs domain exposes grounded, ranked search over your published content
- Gated sites keep working over MCP through an authenticated endpoint, with OAuth 2.1 or machine credentials
- Structured components give models the semantic cues that make chunking clean and retrieval accurate
- Robots directives, sitemaps with real per-page timestamps, and 100/100 Lighthouse SEO
- Analytics separates human views, AI agent views, and bots and crawlers, so you can see who reads you
The MCP server uses the same relevance ranking as the AI Assistant inside your docs, so an agent and a customer get the same answer from the same source.
Knowledge infrastructure, not a scraped website
Being readable by a model is a property of the platform, not a meta tag you add afterwards.
An MCP server on your domain
The Reader MCP server exposes a documentation search tool over hybrid semantic and keyword retrieval. Point ChatGPT, Claude, Cursor, or any MCP client at it and they read your current documentation instead of recalling an old version.
Structure a model can use
Components carry semantic meaning, every heading is anchored and addressable, and each page is available as clean Markdown, so what a model retrieves is content rather than page furniture.
Measured rather than assumed
The Traffic dashboard splits People, AI Agents, and Bots and Crawlers, which turns AI readership into a number you track instead of a claim you make.
The same answer, whether a person or an agent asks
- Hybrid semantic and keyword retrieval, ranked the same way in both surfaces
- Available to ChatGPT, Claude, Cursor, and any other MCP client
- Authenticated endpoints keep gated documentation reachable and still gated
What makes documentation AI-ready
Four things, all generated for you and all switchable.
Generated automatically, with a toggle in Site Config.
Reader for your customers’ AI tools, Authoring for your own team’s.
Authenticated MCP access for private and gated documentation.
Analytics separates people, AI agents, and crawlers on every page.
Frequently asked questions
What teams ask when documentation becomes an interface for agents.
What is llms.txt?
A standard file, much like robots.txt, that helps AI assistants and search engines discover and interpret your documentation. Documentation.AI generates it for you and lets you switch it on or off in Site Config.
Do we need to restructure our content?
No. Using the built-in components is what keeps retrieval clean, and the AI Documentation Agent can apply them to your existing pages for you.
Can AI tools read documentation behind a login?
Yes, with credentials. Public sites use the /_mcp endpoint, and access-controlled sites use /_mcp/auth with OAuth 2.1 for interactive clients or machine-to-machine credentials for automation.
Is the MCP server an enterprise add-on?
No. The MCP server is part of the platform from the free tier, not an upgrade path behind a sales conversation.
Ready to build documentation people and AI agents love?
Live in under 5 min · No credit card required