What Is WebMCP? Why Your Website May Soon Need to Work for AI Agents

Most websites have traditionally been built with two audiences in mind. The people actually using them, and the search engines trying to understand what they contain.

That has influenced the way we build websites for years. Navigation needs to make sense, pages need to be accessible and easy to use, and the content needs enough structure for search engines such as Google to understand what the business does.

AI is starting to introduce another type of visitor: the AI agent.

We've already become used to AI systems reading websites. ChatGPT can find information from them, Google can use them within AI search results and businesses are understandably starting to think about how their content appears in those answers.

AI agents take that idea a step further because they aren't necessarily just looking for information. They can be given a task to carry out on someone's behalf, which could mean checking availability, arranging an appointment or interacting with another system.

That is where an emerging technology called WebMCP gets interesting.

What is WebMCP?

WebMCP is a proposed web standard designed to give AI agents a more structured way of interacting with websites.

At the moment, an AI agent trying to use a normal website often has to do something surprisingly similar to what we do. It needs to understand the page, find the relevant button or form, work out what information is required and then interact with the interface.

WebMCP takes a different approach. A website can explicitly describe certain things that it is capable of doing and make those capabilities available to a compatible AI agent.

If a website allows appointments to be booked, for example, it could expose that as an action and describe the information needed to complete it, such as the date, time, name and email address.

The agent no longer has to work out how the booking page works. The website has provided a structured way for software to interact with that particular part of the business.

WebMCP is still experimental, so this isn't something I think businesses need to start implementing on their websites. What interests me more at this stage is what it tells us about where the web could be heading.

What might this look like in practice?

A hotel is a fairly easy example.

Imagine asking an AI assistant to find a dog-friendly hotel near Keswick for a particular weekend, with parking and a budget of £180 a night. AI can already help find suitable hotels, but checking whether a room is actually available usually means going through the hotel's website or booking system.

If the hotel made availability checking available to AI agents, the assistant could potentially check those dates directly and bring the available options back.

The hotel wouldn't need to give the AI access to its whole booking system. It could make one specific capability available, with its own rules around what information can be requested and what the agent is allowed to do.

The same idea isn't difficult to imagine elsewhere. A restaurant could make table availability accessible, an accountant could make appointment slots available, or a shop could allow an agent to check whether something is in stock.

It doesn't necessarily mean the AI completes the entire process either. There is a big difference between checking that a hotel room is available and actually booking it and taking payment. Some actions could happen automatically while others might still need the person to confirm what they want to do.

The website is only part of the picture

This is where I think WebMCP becomes more interesting from a software point of view.

Many business websites already rely on several other systems behind the scenes. A booking form might connect to a reservation system, an enquiry could be passed into a CRM, an online shop needs access to stock information, and an appointment system may ultimately be reading and updating a calendar.

None of that is particularly new. Businesses have been connecting systems together through APIs and integrations for years.

What is potentially changing is the interface sitting in front of them.

Until now we've mostly expected a person to visit a website and use the interface we've designed for them. WebMCP suggests that an AI agent could become another way of reaching some of the same underlying services.

That doesn't suddenly make every existing website obsolete. People are still going to want to browse a hotel, look at photographs, read about a service or compare products themselves. But some interactions are much more functional. If all somebody wants to know is whether a table is available at 7pm on Saturday, an AI agent doesn't necessarily need to navigate the restaurant's website in the same way they would.

Why the software underneath matters

One of the things I like about this idea is that preparing for it doesn't really require businesses to start adding lots of AI technology.

A business whose website already connects cleanly to its other systems is in a much better position to offer those capabilities through different interfaces in future. If bookings, enquiries, stock or appointments are tied up in awkward manual processes, adding an AI agent on top isn't going to fix the underlying problem.

In that sense, WebMCP is another example of something I've seen quite often with new technology. The newest part gets most of the attention, but whether it is genuinely useful often comes down to much less exciting things such as how the existing software has been designed and how well the different systems work together.

There is value in getting those things right even if WebMCP never becomes widely adopted.

There will need to be limits

There are also some fairly obvious questions around how much an AI agent should be allowed to do.

Letting an agent check whether something is available is relatively straightforward. Letting it make a purchase, cancel a booking or change information held about a customer carries more risk.

Any useful implementation would need clear boundaries around what an agent can access and which actions require confirmation. That is no different in principle from the way we should already design integrations between systems: give something access to what it needs rather than giving it access to everything.

This will become more important if AI agents start carrying out more consequential tasks, particularly where money or personal information is involved.

Is WebMCP something businesses should look at now?

For most businesses, there probably isn't any urgency. WebMCP is still experimental and proposed web standards can change considerably before they become widely supported.

That doesn't mean there's no value in exploring it, though. A business that already has well-defined online processes, such as bookings, appointments, availability checks or product searches, could be a good candidate for experimenting with WebMCP and learning how agent-based interactions might fit alongside its existing website.

I wouldn't build a business-critical process around it yet, but there's a difference between depending on an experimental technology and experimenting with one. For some businesses, getting that experience early could be useful if AI agents do become a more common way for customers to interact with websites.

What interests me most is that the foundations needed aren't particularly new. Websites with well-designed APIs, sensible integrations and clearly defined processes are already in a much better position to experiment with this kind of technology.

Our view

WebMCP is still early, so I don't think every business needs to start changing its website. But for businesses with the right use case, I do think it's worth experimenting with.

The interesting part isn't really whether WebMCP itself becomes the standard everyone eventually uses. It's the wider idea that websites may increasingly be used by software acting on behalf of customers as well as by customers themselves.

Businesses with clear processes, sensible integrations and well-designed software underneath their websites will be in a much better position to explore that shift — whether that means experimenting with WebMCP now or adapting to whatever eventually becomes established.