← Back to digest
Tools & Demos

WebMCP in practice

What's this about?

A hands-on writeup of actually implementing WebMCP on a site, surfacing on HN the same week as OpenAI's WebMCP hackathon from yesterday's digest.

What exactly is WebMCP?

MCP standardises how an assistant talks to tools. WebMCP extends that to websites: a site declares what an agent may do on it, so the agent calls a defined interface instead of clicking around a human interface and guessing. For a shop, that is the difference between an agent simulating a shopper and an agent calling a purchase function.

What's the community saying?

HN 55 points, 55 comments
  1. Interest in concrete implementation details, since most WebMCP coverage so far has been announcements rather than code.
  2. The commercial question that keeps recurring: what does a site gain by making itself easy for agents, if agent traffic does not view ads and does not become a customer relationship?
  3. The comparison to the Accept-header proposal above, as the minimal alternative to a structural one.
← Tailcat: netcat over Tailscale's data planeAWS acquires DuckLabs →