For years, we optimized sites so that a human could read them and a search engine could index them. Today a new question comes up: can an AI also take action on your site?
By take action, we don't mean summarizing your home page. We mean searching for a product, checking availability, filling out a form or booking a slot, without a person clicking at every step.
Reading versus acting
An AI agent can read the HTML of any public page. It understands the text, the headings, sometimes the prices shown. But reading does not mean being able to use your site as a tool.
Without a clear signal, the agent guesses. It may hallucinate an impossible action or give up. WebMCP aims to make that capability explicit: your site announces what it offers as tools to the AI.
How it works, in short
WebMCP relies on an API that the browser exposes to the page, document.modelContext. Your site registers tools there (for example "search the catalogue" or "book an appointment") that the agent can call.
Chrome has offered it since version 149, as a trial (Origin Trial). In practice, two things make a site ready, plus one good practice:
- The trial token: you register your domain for Chrome's WebMCP trial and add the token you receive to your pages. It is temporary: once WebMCP ships for good, it will no longer be needed.
- Declared actions: in JavaScript with document.modelContext.registerTool(), or more simply with a few attributes on an existing HTML form (toolname, tooldescription).
- The good practice: check in your code that the browser supports WebMCP (if (document.modelContext)) before using it, so nothing breaks elsewhere.
Who is it useful for?
If you run an e-commerce site, a booking tool, a comparison site or any service where the end user might one day go through an AI assistant, WebMCP concerns you.
It is not reserved for big tech companies. It's a preparation layer for generative engine optimization (GEO): making your site useful to generative engines and agents, not just visible.
Where to start
Test your URL with WebMCP Check on GEOQuébec. You will see right away what is in place, what is missing, and the starter code to add to your site.
Then read Chrome's WebMCP documentation and plan the implementation with your technical team. The goal is simple: when an AI visits your site, it knows what to do.