Making Your Website Agent-Ready: The Easy Way

For the past couple of years, AI agents have been browsing the web by pretending to be humans. They take a screenshot of the page, ask a vision model to find the button, click, and screenshot again. Each cycle is slow and burns tokens by the screenful, and anything that moves, like an animation or a loading spinner, photographs as noise. Ship a redesign, run an A/B test, drop a cookie banner over the checkout button, and the agent that booked your table yesterday can be lost today.

WebMCP proposes something far more stable: instead of the agent working through rendered pixels, the page just tells it what it can do.

What is WebMCP?

WebMCP is a new web standard, developed by Google and Microsoft and now on the W3C track, that lets a website expose structured tools directly to AI agents. Rather than an agent poking at your interface from the outside, your site publishes a set of tools, each with a name, a description, and a typed schema of inputs. The agent calls a tool with real parameters. The screenshots and the guessing are gone.

From an end user's perspective, the difference is the one that matters: agents get fast and dependable. The agent books the reservation, files the support request, adds the items to the cart, because it's calling a function you defined, not interpreting a picture of your page. And because you decide which tools exist, the site stays in control of what an agent is allowed to touch.

WebMCP landed as an early preview in Chrome 145+ in February 2026, it lives behind a flag or the Origin Trial, and the spec is still moving, with GA expected in Chrome 157. And the cheapest way in is something you probably already have lying around.

The low-hanging fruit: the declarative API

Here's the good news. If your site is built on plain, valid HTML forms, you are most of the way to WebMCP already.

A <form> is a fully-specified contract: these are the fields, these are the values they accept, this is where they're sent. To make it available to agents, all you add is a few attributes describing the intent, and the browser turns the form into a tool on its own.

<form toolname="supportRequestTool"
      tooldescription="Submit a request for support."
      action="/submit">

  <label for="firstName">First Name</label>
  <input type="text" id="firstName" name="firstName">

  <label for="lastName">Last Name</label>
  <input type="text" id="lastName" name="lastName">

  <label for="team">What do you need help with?</label>
  <select id="team" name="team" required
    toolparamdescription="Determines what team this request is routed to.">
    <option value="Customer happiness team">Return my purchase.</option>
    <option value="Distribution team">Check where my package is.</option>
    <option value="Website support team">Get help on the website.</option>
  </select>

  <button type="submit">Submit</button>
</form>

The annotations are three attributes: toolname and tooldescription on the form, toolparamdescription on any field that needs clarifying. Everything else comes from markup you already have: a field without an explicit description gets one from its <label>. The labels you wrote for humans now describe your parameters to agents, too. That's it. The browser converts the annotated form into a tool an agent can invoke.

Not everything is a form

The declarative API buys you exactly what a form can express, and no more. Plenty of actions never touch a form: they live in JavaScript handlers that fetch data from the backend, update state, and re-render. Others need no backend at all, like filtering what's already on the page or recalculating a total from state the browser already holds. Either way, there's no <form> in the path, and nothing to annotate.

Think of the declarative API as the on-ramp: the fastest way to get started with WebMCP. For anything more elaborate, there's the imperative JavaScript API: you register the tool, define its schema, and write the code that runs it.

Where to go next

If you want to play with it, start at the Chrome for Developers WebMCP docs. The declarative API has its own page, and the overview walks through the bigger picture and the early preview program. You'll need Chrome 145+ and a flag flip to try it (unless the website is exposing a specific token to the browser via Origin Trials). Treat it as a preview.

And when you hit the actions no form can express, that's where the imperative API picks up. See you in the next post.