What is WebMCP?
A new browser API that lets AI agents interact with websites the way developers intended. Not through scraping, but through structured tools.
Explore WebMCP
The State of WebMCP
Where adoption actually stands right now, and what's still an open question
How WebMCP works
Practical guide to making your site agent-ready
WebMCP vs browser automation
Why scraping is the wrong approach for AI agents
WebMCP vs MCP
Two different protocols, two different layers
Lighthouse agentic browsing score
How Chrome now scores your site for AI agents
Browser support
The Chrome origin trial, polyfill coverage, what works today
Implementation examples
Copy-paste registerTool() patterns for common use cases
WebMCP for e-commerce
Making Shopify and retail sites agent-ready
schema.org markup as a WebMCP shortcut
Most sites already have the JSON-LD a read-only tool needs
The problem with how AI agents use the web today
Right now, when an AI agent like Claude or ChatGPT needs to interact with a website, say to book a flight, add something to a cart, or fill in a form, it has two bad options:
Scrape the page
The agent reads your HTML as raw text and tries to figure out what to click. Brittle, slow, breaks on every design update, and blocked by most anti-bot systems.
Use a custom API integration
The agent has a hand-crafted plugin for your specific site. Only works for the biggest platforms, and requires constant maintenance.
Neither approach scales. The web has billions of sites. AI agents can't have hand-crafted knowledge of all of them.
WebMCP: a native standard for agent-site communication
WebMCP (Web Model Context Protocol) is a browser API, initially introduced by Google in Chrome, that lets websites expose structured tools directly to AI agents.
Instead of an agent guessing what your "Add to Cart" button does, your site can explicitly declare:
{
name: "addToCart",
description: "Add a product to the shopping cart",
parameters: {
productId: "string",
quantity: "number"
}
}The AI agent sees this declaration and knows exactly how to call addToCart. No guessing, no scraping.
Scraping vs. calling tools, side by side
Same task, same page. On the left an agent has to read the DOM and guess. On the right the page declares what it can do, and the agent just calls it.
How it works
Site exposes tools via document.modelContext
A JavaScript API in the browser lets any page register named tools with descriptions and input schemas. It's part of the browser. No external service needed.
AI agent discovers the tools
When an AI agent navigates to a page, it checks document.modelContext for available tools, the same way a developer checks an API spec.
Agent calls the tool directly
The agent calls the tool with structured arguments. The tool's execute function runs in the browser, doing exactly what the site owner intended.
Result returned to the agent
The tool returns structured data back to the agent, which can continue the conversation or take further actions.
Browser support
Native document.modelContext support is running as a public origin trial in Chromium 149–156. The Aigentably snippet loads the WebMCP polyfill automatically for all other browsers, so your tools work everywhere today.
| Browser | Native WebMCP | Via polyfill |
|---|---|---|
| Chrome 149–156 | ✓ Origin trial | ✓ |
| Edge 149–156 | ✓ Origin trial | ✓ |
| Chrome < 149 | — | ✓ |
| Firefox | — | ✓ |
| Safari | — | ✓ |
| Brave / Opera | — | ✓ |
The document.modelContext API
WebMCP exposes a single browser API. Calling document.modelContext.registerTool() declares a tool the AI agent can discover and call.
document.modelContext.registerTool({
name: "searchProducts",
description: "Search the product catalogue",
inputSchema: {
type: "object",
properties: {
query: { type: "string" }
},
required: ["query"]
},
execute: async ({ query }) => {
// runs in the browser
return window.searchAPI(query)
}
})Lighthouse now scores WebMCP
Lighthouse 13.3+ includes an agentic browsing category that measures how accessible your site is to AI agents. It checks whether your site exposes WebMCP tools, validates their schemas, and audits accessibility-tree consistency.
Sites without WebMCP tools score 0 in this category. As AI agent traffic grows, this score will matter the way mobile performance scores matter today.
Why it matters
Think about what SEO did for search engines: it gave site owners a way to communicate structure and intent directly to crawlers. WebMCP does the same thing for AI agents.
Sites with clear WebMCP tools are more useful to AI agents. That means they get recommended, used, and acted upon more than sites that don't have them.
It's early. The standard is new. But the direction is obvious: being agent-ready will eventually be as expected as being mobile-ready.
Further reading
Make your site agent-ready
Aigentably handles the WebMCP infrastructure. You define the tools.
Get started free