AI Native Documentation

Most documentation traffic now comes from AI crawlers and agents read docs more than humans do. Here’s how teams are adapting.

Render Markdown for Agents

Agents send Accept: text/markdown in the request header when fetching pages. Servers can detect this and return markdown instead of HTML. Less tokens, cleaner content.

Bun and @thdxr shared this pattern.

AI Prompts

Docs now include AI prompts you can paste directly into your agent to handle setup.

Examples:

• Clerk Docs
• Sentry Docs
• Supabase Docs

Ask AI

Docs sites embed AI chat as a sidebar or popup. Many are powered by Kapa.

Examples:

• Cursor Docs
• Prisma Docs
• Vercel Docs

Copy Page as Markdown

Make it one click to copy documentation as markdown.

Cursor and Vercel do this with a “Copy page” link in the sidebar.

Markdown URL Support

Support viewing pages as Markdown by adding .md to the end of the URL. nuqs does this and includes a “Copy Markdown URL” button along with “Open in…” options to open the markdown URL directly in ChatGPT or Claude.

llms.txt

Put a llms.txt in the root. Some sites include links to key docs, others include the full content in markdown format.

Examples:

• nextjs.org/docs/llms.txt
• tanstack.com/llms.txt
• ui.shadcn.com/llms.txt

Some sites provide both llms.txt and llms-full.txt files.

MCP Servers

Projects run their own MCP servers for documentation, like Astro Docs MCP Server. Context7 works across documentation for many tools and libraries.