WebMCP is now something you can actually run. Chrome ships it as an early preview, and DevTools has a panel for inspecting the tools a page registers. For a proposal that was announced in February, that is fast.
The short recap, if this is new to you: WebMCP lets a page hand an agent a set of tools through navigator.modelContext. Real functions with names, descriptions and input schemas. Search this catalogue, add this to the basket, book this slot. Instead of an agent screenshotting your page and guessing where to click, it calls the function you gave it. The longer explanation is in What is WebMCP.
What shipping in Chrome changes is not adoption. Almost nobody's site exposes tools yet, and the spec is still a W3C community group document that will keep moving. What it changes is that the loop is now closable: you can register a tool, open DevTools, and watch an agent-shaped thing call it. Before this, WebMCP was a document. Now it is a debuggable surface.
It also means the checker ecosystem arrived early. There are already browser extensions that score pages on WebMCP readiness, and a directory or two of sites that expose tools. The pattern from UCP is repeating a protocol gets a spec, then validators, then a market view, and the sites themselves come last.
If you run an interactive site, the one thing worth doing this month is small: pick your single most obvious action and expose it as one tool. Not a full catalogue. One search function with a clear description. It costs an afternoon, it puts you ahead of essentially everyone, and it gives you something real to test when the agents that speak WebMCP start arriving.
We wrote a walkthrough for that afternoon: how to expose WebMCP tools on your site. And the check now reads registered tools as part of its pass, so you can see what an agent sees once yours is live.
