Everything the platform does, in one page
An AI agent that drafts with you, workflows that keep every page current after you stop looking, publishing on your own domain with the access model you need, and an MCP server so your docs are as readable to AI tools as they are to people.
The world's best product teams trust Documentation.AI
From next-gen start-ups to established enterprises.
Four things, on one platform
Not four products. The same project, the same content, and the same navigation underneath all of them.
- Write
- An agent in the editor with your repositories, issues, and wiki as live context, plus MCP authoring from Cursor or Claude Code.
- Self-Updating Docs
- Nine workflow templates on merges and schedules, with analytics naming the pages that are failing rather than the ones that are old.
- Publish
- Generated API references, three access modes, versions and languages, on your own domain at 100/100 Lighthouse.
- AI Access
- A citing assistant in your docs, a Reader MCP server for your customers’ tools, and llms.txt generated for you.
An agent that starts with your context, not an empty file
The AI Documentation Agent drafts, rewrites, and reformats inside the editor, using your own components and site configuration rather than generic Markdown.
- GitHub, GitLab, Jira, and Confluence connect as live context sources, queried at runtime through your OAuth connection instead of imported as a snapshot that starts going stale immediately
- A visual web editor with slash commands for anyone who does not want Git, and Markdown and MDX in your own repository, with a preview build on every pull request, for anyone who does
- The Authoring MCP server lets Cursor or Claude Code inspect files, update pages, manage branches, and deploy through your normal workflow
- Attached PDFs, documents, and spreadsheets are read and converted, so existing material becomes pages rather than a retyping project
Everything the agent produces arrives as a commit you review. Nothing it writes publishes itself.
Maintenance that runs whether or not anyone is watching
Nine workflow templates cover broken links, grammar, style guides, navigation, changelog generation, and code-aware updates, triggered on a merge or on a schedule.
- Each run returns a summary, a report, and a commit, so drift arrives as a proposed edit rather than as a support ticket
- Analytics covers traffic, search behaviour, page ratings, and the comments readers actually write
- The questions Ask AI answered with low confidence name the pages worth fixing, which is a different list from the pages that are simply old
- Connectors watch the repository and issue tracker your team already uses, so a change in the product is a change proposed to the page describing it
The point is not that the docs update themselves. It is that nobody has to remember to check.
Your domain, your access model, no build to tune
Point a navigation group at an OpenAPI file and the reference generates itself, playground included. Point one page at a single operation when you want to write the rest by hand.
- Parameter documentation, request examples in cURL, Python, JavaScript, Go, and Ruby, response examples for every status code, and a live playground that sends real requests
- Public, partial, and private access, with a password, JWT, or OAuth 2.0. JWT and OAuth both carry per-user roles, so each reader sees the pages their role allows
- Products, versions, and languages as navigation dimensions, each able to carry a landing page of its own
- Your own domain or a subpath, 100+ components, and 100/100 Lighthouse Performance, Accessibility, and SEO with no performance budget for anyone to defend in review
Content is standard Markdown and MDX in a Git repository. Migrations from GitBook, Docusaurus, ReadTheDocs, and custom static sites are supported.
Read by the tools your customers already have open
A model with a training cutoff answers from the version of your product it happened to see. Everything here exists to point it at the current one instead.
- The AI Assistant answers inside your docs and cites the exact page and section behind every answer, so a reader can check it rather than open a ticket to confirm it
- The Reader MCP server gives ChatGPT, Claude, Cursor, and any MCP client grounded search over your published documentation, including access-controlled sites
- Auto-generated llms.txt and clean per-page Markdown, so retrieval lands on the section that answers the question instead of guessing across pages
- Hybrid search combining semantic understanding with keyword matching, ranked with a heading-level boost, for people and for machines alike
Public sites use the /_mcp endpoint. Access-controlled sites use /_mcp/auth, with OAuth 2.1 for interactive clients and machine-to-machine credentials for automation.
What every plan starts with
Published pricing, no procurement cycle, and nothing on this page gated behind a sales conversation.
From signup to a live documentation site, with no credit card required.
Lighthouse Performance, Accessibility, and SEO on every page you publish.
Templates for links, grammar, style, navigation, changelogs, and code-aware updates.
Reader for the AI tools your customers run, Authoring for your own team.
Go deeper in the documentation
Every capability above, written up properly, with the configuration and the limits. The same four columns the Product menu holds.
Write
Self-Updating Docs
Publish
Frequently asked questions
The questions people ask while comparing platforms.
What is Documentation.AI?
A documentation platform with an AI agent built into it. You write in a visual editor or in Markdown and MDX in your own Git repository, background workflows keep pages current as the product changes, and the published site is readable by people, by search engines, and by AI tools through llms.txt and MCP.
Do we have to use Git?
No. The web editor provides branches, previews, and publishing from the browser, with no repository involved. Teams that do use Git get Markdown and MDX in their own repository with a preview build on every pull request. Both work on the same project at the same time.
Can it generate an API reference from our OpenAPI spec?
Yes. Point a navigation group at the OpenAPI file and the reference generates, with parameter documentation, request examples in cURL, Python, JavaScript, Go, and Ruby, response examples for every status code, and a live playground that supports the authentication schemes declared in your spec.
What can we put behind authentication?
A whole site or individual pages. Public, partial, and private modes are supported, with a password, JWT, or OAuth 2.0, and connect to Okta, Auth0, Azure AD, or any OAuth 2.0 or OIDC provider you already run. JWT and OAuth carry per-user access roles.
What does the AI agent actually change without us asking?
Nothing that publishes. Workflows run audits and propose updates, and every result arrives as a summary, a report, and a commit for review. The agent drafts and rewrites inside the editor when you invoke it.
How do we move what we already have?
Content is standard Markdown and MDX, and migrations from GitBook, Docusaurus, ReadTheDocs, and custom static sites are supported. The agent can also read attached PDFs, documents, and spreadsheets to convert material that was never in a docs tool.
Ready to build documentation people and AI agents love?
Live in under 5 min · No credit card required