When websites start explaining themselves to AI agents (and the buttons finally stop being guesswork)
Websites were designed for people. We can look at a booking form and usually work out what it wants, even when the labels are not especially helpful. An AI agent has to interpret the same page from its text, structure and controls.
That works reasonably well until two buttons both say “Submit”, or a form hides the next step until an earlier field has been completed. The agent then has to guess what the designer meant. This is manageable when it is looking up opening times. It is less reassuring when it is placing an order.
WebMCP is an experimental proposal intended to make those tasks clearer.
What WebMCP does
WebMCP would let a website describe useful actions in a structured way. Rather than making an agent inspect a page and work out how to book an appointment, the site could provide a tool called “find available appointments” and explain what information it needs.
A shop might provide tools for searching products, checking stock and adding an item to a basket. A service website might let an agent request a quotation or check the progress of an existing job.
The website would choose which actions to make available. Each action could have a clear description and defined inputs, so the agent does not have to infer the meaning of every field and button.
The ordinary website would still be there. People could continue to browse and use it as they do now. WebMCP would add another way for compatible browser agents to complete the tasks the site has deliberately exposed.
Why this may be more reliable
Browser agents can already use many websites. They read the page, follow links, fill in forms and respond to what appears next. The difficulty is that small changes can upset the process.
A renamed field, a redesigned menu or an unexpected pop-up may be enough to send the agent down the wrong route. People often recover because they understand the wider context. An agent may continue with a choice that looks plausible but is not what the user asked for.
Structured tools remove some of that ambiguity. The agent knows the name of the action, the information it must supply and the sort of result it should receive. It is much the same reason a clear form is easier for a person to use than one assembled from mystery boxes.
This could be particularly useful for longer processes such as travel bookings, service applications and insurance claims. These often involve several stages and questions that change according to earlier answers. A structured action can return a clear error when information is missing instead of leaving the agent to look for whichever field has quietly turned red.
Useful jobs for businesses
Many everyday business tasks are quite repetitive. Customers ask whether an appointment is available, whether an item is in stock or whether a service covers their area. A browser agent could handle some of that work if the website gives it a dependable way to ask.
It might also prepare an action without completing it. An agent could build a basket, gather the details for an enquiry or find suitable appointment times, then ask the person to check everything before it is submitted.
That is likely to be more useful than handing the whole process over without review. People generally want help with the tedious part, not a surprise purchase made on their behalf.
It does not remove permissions or security
Describing an action does not mean every agent should be allowed to carry it out. A user may need to sign in, and the website still has to check what that account is permitted to do.
Reading delivery options is different from changing an address. Preparing an order is different from paying for it. Expensive, sensitive or irreversible actions should still require clear confirmation from the person involved.
The website also needs to validate the information it receives. Agents can misunderstand instructions, and web pages can contain misleading or malicious content. A structured tool gives the agent a clearer route, but it does not make the rest of web security optional.
Good implementations will keep tools narrow and make important consequences obvious. If an agent is about to cancel a booking or spend money, the confirmation should say exactly that. “Continue” has done enough damage already.
How current is it?
Chrome announced a WebMCP origin trial in June 2026, associated with Chrome 149. An origin trial allows developers to test an experimental browser feature on real websites for a limited period.
WebMCP is not a finished standard and it is not supported across every browser. The details may change as it is tested. Businesses do not need to rush out and add it to their websites now.
It is still worth watching because browser agents are becoming more capable, and people are beginning to use them for practical tasks rather than simple questions. If that continues, websites may need to explain their useful actions to software as well as presenting them to people.
The two should support each other. A clear website remains important because people still need to understand their choices and decide whether they trust the business. An agent-friendly layer may make the practical steps easier, but it should not make the person feel they have lost control of them.
More Tech Trends
If you enjoyed this article, we have plenty more in Tech Trends. We look at new technology, what it is useful for and where its limits are.
Here are a few to get you started:
No Internet, No Problem: The New Software That Works Perfectly Offline
What Cursor is actually doing when it rewrites your code (and why it feels uncomfortably accurate)
The App Store Shortcut: How "Hybrid" Websites are Replacing Phone Apps
Why we write about technology
We build websites and software, so keeping up with changes in technology is part of the job. We need to know what is useful now, what may be useful soon and what can safely be left to enjoy its moment on the internet.
That knowledge shapes the things we build. Here are a few examples of our work:
TrailTrack - The Best Ever Outdoor Hiking Website
If you would like us to build something
If you need a website, custom software or a better way to manage a process in your business, get in touch and tell us what you need.
What is WebMCP?
WebMCP is an experimental proposal that lets websites describe their useful actions in a structured way so AI agents can use them more reliably. Instead of an agent guessing what a button does, the website explains actions like 'find available appointments' with clear descriptions and required inputs.
Can I use WebMCP on my website right now?
Not really — it's still experimental. Chrome announced an origin trial in June 2026 for developers to test it, but it's not a finished standard and isn't supported across all browsers yet. You don't need to rush to add it now.
Will AI agents be able to make purchases on my behalf without asking?
They shouldn't. Good implementations require clear confirmation before expensive, sensitive, or irreversible actions. An agent might build a basket or gather details, but you should always review and approve before anything is submitted or paid for.