AI Agents (WebMCP)

The site can serve notes not only to people but to AI agents too — in a structured way, no HTML scraping required.

What it is

WebMCP is an emerging web standard (W3C draft): a page registers "tools" in navigator.modelContext, and an AI agent in the browser can call them. markdown-publish registers four:

Tool	What it does
search_notes(query, limit)	Search across notes: title, link, snippet
get_note(slug)	The note's full markdown + backlinks
list_notes()	A list of all the site's pages
get_backlinks(slug)	Who links to a note

If the browser has no native navigator.modelContext, the site installs its own lightweight polyfill — no third-party dependencies.

Try it yourself

Open DevTools (F12) → Console:

await navigator.modelContext.callTool({
  name: 'search_notes',
  arguments: { query: 'configuration', limit: 3 },
})

And for crawlers too

• /llms.txt — a site map for AI crawlers (llmstxt.org).
• Every page is prerendered into clean HTML — readable without JavaScript.
• sitemap.xml, robots.txt, OG cards and JSON-LD — regular SEO is covered too.

Hidden notes (buildMode: public, see Configuration) are not served to agents — they simply don't exist in the build.