Teaching your website to talk to AI agents
WebMCP is an emerging web standard that lets your site tell an AI agent exactly what it can do, instead of leaving it to guess from your markup. Here's why it's worth getting ahead of.
Over the past couple of years, AI assistants have become remarkably good at browsing the web on our behalf — booking appointments, comparing prices, filling in forms, checking order status. The trouble is, most of them do this the hard way: by taking a screenshot of your page, guessing which button says “Checkout,” and clicking around much like a person would, only clumsier. It works, but it’s slow, it’s fragile, and it breaks the moment you redesign your navigation.
WebMCP is a new, emerging web standard — backed by Google and being developed openly through the W3C’s Web Machine Learning Community Group — that fixes this by giving websites a proper way to say, in plain terms an AI can understand: “here’s what you can do on this page, and here’s exactly what I need from you to do it.”
Think of it as a menu, not a maze
Right now, when an AI agent lands on your site, it has to work everything out by trial and error — reading your HTML, spotting a form, hoping it has correctly identified the “quantity” field rather than the “postcode” field. WebMCP lets a website skip all that guesswork by publishing a short, structured list of the actions it supports — “check availability,” “request a quote,” “add to basket” — along with the exact information each one needs. The agent reads that list, much like reading a restaurant menu instead of interrogating the kitchen, and calls the action directly.
Under the bonnet, this happens through a small addition to the browser itself: a new navigator.modelContext interface that a site’s developers can use to register these “tools,” or, more simply, by adding a couple of extra attributes to an existing HTML form. Either way, the visitor’s experience on screen doesn’t change at all — WebMCP simply opens a second, machine-readable door alongside the one your human visitors already use.
It’s also worth being clear about what WebMCP isn’t. You may already have heard of MCP (the Model Context Protocol), which lets AI tools like Claude or ChatGPT connect to backend systems and services. WebMCP is the browser-side sibling of that idea — built specifically for the live, interactive page a person is looking at, rather than for systems working away in the background. The two are designed to work well together, but WebMCP is squarely about what happens inside the browser.
Why it’s worth getting ahead of
WebMCP is still early — Chrome currently offers it as an experimental preview, and the standard itself is subject to change. But the direction of travel is clear, and there are good reasons for organisations to start thinking about it now rather than later:
Reliability
Agents that call a defined tool don’t break when you tweak your layout, rename a button, or run an A/B test. Compare that to screen-scraping approaches, which fall over with almost every redesign.
Visibility to a growing channel
As more people delegate everyday tasks — price comparisons, bookings, research, form-filling — to AI assistants, sites that are easy for those assistants to use correctly are more likely to be the ones that get chosen, understood accurately, and transacted with.
Control, not exposure
Because you define exactly which actions are available and what data they expose, WebMCP gives you a deliberate, well-scoped interface — the opposite of leaving agents to infer things from your markup, which can misfire in ways you don’t control.
No duplicate API to maintain
Rather than building and maintaining a separate public API purely for automation, WebMCP lets you expose the same actions your site already performs, described in a format agents can read directly.
For anything customer-facing — a booking system, a product catalogue, an account portal, a quote request form — that’s a meaningful head start on a web that’s increasingly being browsed by more than just people.
Where to start
You don’t need to overhaul your website to benefit from WebMCP. For most organisations, the sensible first step is identifying the handful of actions on your site that are genuinely worth exposing to AI agents, and planning how those would be described. It’s exactly the kind of practical, incremental AI work we help clients with every day.
If you’d like to talk through what WebMCP could mean for your website or web application, contact us today — we’d be happy to help you work out where to start.
