Question 1

What is WebMCP?

Accepted Answer

WebMCP is a proposed web standard that lets a website give AI agents a list of actions they can use directly, like searching products, booking a table or filling in a form. Instead of trying to click buttons and figure out a page like a human, the agent uses the actions the website provides. Engineers from Google and Microsoft are developing it in the W3C’s Web Machine Learning Community Group. Each action is a tool with a name, a description and the inputs it needs. A site can turn an existing HTML form into a tool with a few attributes, or register a JavaScript function through the browser’s document.modelContext API. There are two main ways people use it. In co-browsing, you and an AI assistant work on the same page in your browser. ChatGPT’s desktop app already does this with WebMCP tools. Gemini in Chrome is the same kind of assistant, but it can’t use WebMCP tools yet. In remote browsing, an AI app opens its own browser and completes the task for you, with or without you watching. Because the website tells the agent exactly what it can do, tasks are faster, more reliable and more predictable than screen scraping. webmcp.com lists 1,329 websites that already support WebMCP.

Question 2

How do I make my website work with AI agents like ChatGPT?

Accepted Answer

Give agents tools instead of making them guess. With WebMCP, your site tells an AI agent exactly what it can do, like search, add to cart or book. ChatGPT’s desktop app already uses these tools in its built-in browser. There are two ways to add them. If your site uses plain HTML forms, give each form a name and a description with the toolname and tooldescription attributes, and the browser turns it into a tool. For anything a form can’t do, register a JavaScript function with document.modelContext.registerTool(). No tools yet? Use “Build WebMCP tools for your website” at the top of webmcp.com. Your coding agent can write them, or you can scan your site and get a proposed set to review.

Question 3

Which AI assistants and browsers support WebMCP today?

Accepted Answer

As of October 2026, ChatGPT is the main AI assistant that uses WebMCP tools: its desktop app has done so in its built-in browser since August 25, 2026. Brave’s assistant, Leo, can use them in Brave Nightly behind a flag. Gemini in Chrome can’t use WebMCP tools yet, and Claude for Chrome doesn’t use them. Chrome and Edge are testing WebMCP in an origin trial, Brave has it behind a flag, and Firefox and Safari have no implementation. Developer tools like Stagehand and Vercel’s agent-browser already support it.

Question 4

How is WebMCP different from MCP or llms.txt?

Accepted Answer

MCP connects an AI app to servers and services somewhere else; you can think of it as an API for AI. WebMCP lets the web page itself hand tools to the AI agent that’s using it, in the browser and inside the user’s own session. llms.txt is a text file that helps AI read your site, while WebMCP lets AI use it. The stated goal of the WebMCP proposal is co-browsing, where users stay on a website and work with it alongside their AI assistant, rather than the AI operating in a separate application. In many cases, this is better for everyone. Users can get through complex interfaces more easily while staying in control, and website owners keep users engaged on their own site instead of moving the experience into a headless browser. WebMCP can also be used from remote automated browsers. Website owners can block that kind of access with their existing bot protection or other access controls.

Question 5

Can AI agents buy or book things on my website?

Accepted Answer

Yes, if your site gives them tools for it. With WebMCP, an agent can search, fill a cart or start a booking by calling actions your site defines. Anything that spends money or makes a commitment should ask the person to confirm first. Shopify and Stripe already have this built in. Shopify stores share WebMCP tools for search, products and the cart, plus checkout tools that place an order only after the buyer confirms; we track more than 2.3 million Shopify stores. Stripe Checkout has its own WebMCP tools for paying. Other sites add their own tools, and the webmcp.com directory shows how shops and travel sites do it.

Question 6

How do I check if AI agents can use my website?

Accepted Answer

AI agents can already try any website by reading the screen, but that’s slow and breaks easily. To see whether your site gives them tools they can call directly, scan it with “Check and improve your WebMCP implementation” at the top of webmcp.com. The scan lists the WebMCP tools it finds and grades how clearly they’re defined. If it finds tools and your site isn’t in the WebMCP directory yet, the scan also sends a listing request, and we review it before the site is listed.

The WebMCP Directory.

Validate your tools

Build with a coding agent

Scan your website

Let the agent close the loop.

1. Give your agent the prompt

2. Agent shares WebMCP implementation with review service

3. Agent gets a grade + feedback

4. Agent fixes and resubmits

Shopify stores

Find a Shopify store

Stripe Checkout

Frequently Asked

For Agents