⏺ webmcp-ai.dev Get the audit →

home / explainer

What is WebMCP?

WebMCP is a W3C proposal from Google and Microsoft that lets websites expose their functionality as callable tools for AI agents via document.modelContext. Here's how it works, its current browser support, and why it matters for your site.

Updated 2026-08-28 · webmcp-ai.dev research desk

WebMCP (Web Model Context Protocol) is a proposed web standard that lets a website hand AI agents a set of structured, callable tools — instead of forcing them to screen-scrape the page and simulate clicks. It was jointly proposed by engineers at Microsoft and Google, published as a unified proposal in August 2025, and is now a draft of the W3C Web Machine Learning Community Group.

The pitch is simple: agents are already browsing the web — Gemini in Chrome, browser extensions, automation frameworks. Today they operate your site the hard way, by looking at pixels and guessing which button does what. WebMCP lets your page say, in a machine-readable contract: "here is what you can do here, and here is exactly how to do it."

How WebMCP works

A page registers tools through the document.modelContext API (early articles and demos used navigator.modelContext; the getter moved to Document, and Chrome 150+ deprecates the Navigator alias). Each tool has a name, a natural-language description written for the model, a JSON Schema describing its inputs, and an execute callback that runs your existing page JavaScript: 

document.modelContext.registerTool({
  name: "search_products",
  description: "Search the store catalog. Returns product names, prices and URLs.",
  inputSchema: {
    type: "object",
    properties: {
      query: { type: "string", description: "What the user is looking for" },
      maxResults: { type: "number" }
    },
    required: ["query"]
  },
  annotations: { readOnlyHint: true },
  async execute({ query, maxResults = 5 }) {
    const results = await storeSearch(query);
    return results.slice(0, maxResults);
  }
});

When an agent-capable browser visits the page, it discovers the registered tools, reads their schemas, and can invoke them — with the user in the loop, inside the user's existing session. That last part is the quiet superpower: the agent acts as the logged-in user, with no API keys, no OAuth server, and no separate backend integration.

There is also a declarative path: annotate a standard HTML form with tool attributes and the form itself becomes an agent tool with zero JavaScript. Chrome's "Le Petit Bistro" demo exposes a restaurant reservation form to agents this way.

Browser support in 2026

WebMCP is not the same as MCP

Anthropic's Model Context Protocol (MCP) connects AI assistants to backend servers, out-of-band from your website. WebMCP runs inside the page, in the user's browser and session. They're complementary layers of the same "agentic web" stack — we break the whole stack down in WebMCP vs MCP vs llms.txt.

Why it matters for your business

Early research on the browser-tab-as-tool-server approach reports agents completing tasks with dramatically fewer tokens and higher success rates than pixel-based actuation (one 2025 paper measured a 67.6% token reduction and ~98% task success on tool-based flows). Meanwhile Cloudflare scanned 200,000 sites for agent readiness — only about 4% passed its content-accessibility check. Consumer brands including Expedia, Shopify, and Target have been reported among early WebMCP experimenters. The gap between agent-ready sites and everyone else is the point: when an agent can't use your site, it completes the task on a competitor's.

Which tools should your site expose? That depends on your business type — we maintain a cookbook of WebMCP tool patterns by industry, and our deep audit produces a blueprint specific to your site.

Sources and further reading