What is the WebMCP API?

Current site processing for AI agents

For LLMs and other AI agents to interact with your website, they need to visit the site, screenshot it, parse HTML, and click buttons around. Kinda like a human, but clunkier and slower — albeit automated.

It's a slow and expensive process, since all of that DOM content eats up LLM tokens. And it's fragile, because if the layout ever changes, the agent breaks.

Where WebMCP comes into play

The new web protocol, which is currently a proposal, enables AI agents to talk directly to your website through a structured interface. Rather than the AI needing to parse your entire DOM structure and guess how to use your site, it provides the ability for your site to explicitly declare what it is capable of.

It does this by publicly exposing a set of tools, parameters, and instructions, which tell LLMs how to interact with your site.

WebMCP fits into what you already have

Think for a second about how responsive design works. You didn't need to rebuild your entire site for different viewports, but rather, you just added in some media queries and updated the CSS. The overall concept is pretty simple, and provided the ability for your site to be fully responsive.

WebMCP is very similar. All you need to do is add a few HTML attributes to existing DOM elements and forms, or use a few lines of JavaScript to register your new tools. And that's it — your site now has a direct line of communication with any AI agent that needs to interact with it.

WebMCP is a type of MCP

Model Context Protocol, or MCP, provides this same ability, but requires a completely separate server to expose the proper tool calls and instructions. WebMCP follows the same concept, but makes your web page the server. No separate infrastructure needed, as the implementation is just HTML attributes or JavaScript.

Since all of the tooling and instructions live directly in the browser and use the existing user session and authentication process, no additional infrastructure is required to implement WebMCP.

WebMCP status and how to enable it

WebMCP is currently an early proposal to expose this new web browser API, which offers this ability to communicate with LLMs. It's currently only available in Google Chrome Canary, and is currently behind a separate flag that needs to be activated.

You can activate it by opening up Chrome Canary version 146 or higher, typing in chrome://flags, and looking for the “WebMCP for testing” setting. Then just change the status to “Enabled” and relaunch the browser. This will allow you to start testing WebMCP in its experimental state.

Make no mistake — WebMCP is going to be an extremely important part of the web, and tells you a bit about where the future of it is heading. It adds the ability to make websites fully AI-native, rather than being hacked around to expose this tooling.

And the implementation is surprisingly simple.