WebMCP: the new web standard that lets AI agents use websites, from search to checkout

WebMCP is a new web standard that lets a website tell AI agents directly which actions they can perform. With WebMCP, an agent can walk into an online store, search for a product, pick the right variant, add it to the cart and take you all the way to checkout. Without hunting for buttons on the page.

Shopify has already enabled it on every Liquid store, Chrome has been testing it since June 2026, and the built-in browser in the ChatGPT desktop app already supports it.

How did an AI agent browse a website before WebMCP?

Until now, an AI agent browsing a website did roughly what we do. It looked at the page, worked out where a button was, clicked it, filled in a field and hoped everything worked.

The problem with this approach is that a website’s interface is designed for human eyes. An AI agent has to interpret it every time, and a changed layout or a moved button is enough to make it fail.

How does WebMCP work?

WebMCP changes the mechanism. The website tells the agent: "These are the actions you can take". For example:

• search for a product
• fill in a form
• add a product to the cart
• book an appointment

Technically, the website exposes these actions as tools through JavaScript. Each tool tells the agent its name, what it does, what information it accepts and which action it performs.

In practice, a website with WebMCP has a second interface, built in code and designed for AI agents instead of humans. The agent calls the right tool with the right data, instead of guessing where to click.

The name echoes MCP, the Model Context Protocol, the standard AI agents use to connect to external tools. The difference is that with WebMCP the tools come from the web page open in your browser, with no separate server or connector to set up.

Who already supports WebMCP?

WebMCP is still early, but three major players have already moved:

1. Shopify has enabled WebMCP tools on every Liquid storefront, with nothing for merchants to install. According to Shopify’s changelog, an agent can search the catalog, read products and variants, manage the cart and proceed to checkout.
2. Chrome opened a WebMCP origin trial starting with Chrome 149, announced on June 9, 2026.
3. ChatGPT supports WebMCP in the built-in browser of its desktop app: OpenAI calls this feature site tools.

Most websites today are not yet ready to be used by AI agents.

Is WebMCP only for online stores?

WebMCP applies to any web app, not only online stores. A web app can expose tools to create a project, edit a document, book something, generate a report or run any other action the application supports.

If you use Shopify, part of the work is already done. If you run your own web app, you need to add the tools, and this is where it gets interesting.

How do you implement WebMCP on an existing web app?

If you already have a web app, implementing WebMCP is much simpler than you think. Open Claude Code, Codex or another agent inside the project and ask something like:

Analyze this web app and implement WebMCP tools so an agent can perform these actions.

Then describe what you want to make available to agents, for example search, creation and editing. Claude Code or Codex implements the tools you need.

The step you can’t skip is the review: you have to decide carefully which actions AI agents are allowed to take. A tool that reads a catalog carries different risks from a tool that changes data or completes a purchase. It’s the same reasoning behind the security checks to run on an app built with AI before launch.

If it’s your company’s web app and you want to work out which actions to expose and how, that’s the kind of work I handle in custom software development projects.

Why start with WebMCP now?

In my view, this is the part many people are underestimating. Over the next few years, a growing share of internet traffic will come from AI agents doing the same things a user would: buying products, booking calls, submitting requests.

Preparing a website for AI agents today is like optimizing for Google back when nobody had yet understood how much SEO mattered. We are starting to build websites with two interfaces: one for humans and one for agents.

Whoever starts building the second interface today will be ahead when agents really start browsing the web on our behalf. Now is the moment to try it on your web app.