Skip to main content

​
AI-native Docs

This documentation is designed to work well for both human readers and AI tools. The repo now includes the machine-readable and navigation-level pieces that Mintlify can use directly from the docs/ source directory, while the hosted site can layer on Mintlify-managed features like Markdown export and MCP access.

​
Reading

The docs now enable Mintlify’s contextual menu through the docs-root docs.json config. That gives readers one-click ways to:
  • copy the current page as Markdown
  • open the current page as Markdown
  • send the page to ChatGPT, Claude, Perplexity, or Windsurf
  • copy or install the hosted MCP endpoint in supported tools
The command reference and workflow guides are already organized for quick AI retrieval. For example:

​
Writing

This repo keeps its AI-facing capability summary in the docs-root skill.md. That file tells agents what kagi can do, which inputs matter, and which constraints apply. For editor-native writing workflows, the repo now also ships:
  • CLAUDE.md for Claude Code
  • .cursor/rules/kagi-docs.mdc for Cursor
  • .windsurf/rules/kagi-docs.md for Windsurf
The contributor-facing source of truth for doc maintenance remains: Those pages explain how to update user-facing docs when behavior changes, which is the maintenance loop this project currently keeps under version control.

​
Discovering

This repo now ships a docs-root llms.txt alongside skill.md so agents can discover the documentation surface from the Mintlify source directory without guessing. When the site is deployed on Mintlify, the hosted docs can additionally provide:
  • Markdown-rendered pages via the .md URL suffix
  • a hosted MCP server for up-to-date documentation search
  • semantic search and related-page recovery on 404s
Because those features are deployment-level behavior, keep claims about them aligned with the live site rather than assuming they are active in every environment.

​
Repo Contract

The documentation should stay AI-friendly by default:
  • keep page titles and descriptions specific
  • keep command examples copy-pasteable
  • prefer stable links between guides, commands, and reference pages
  • update llms.txt and skill.md when the public docs surface changes in a meaningful way