Intelligence. Built into the DOM.
WebMCP (Web Model Context Protocol) is an open web standard proposed under the W3C Web Machine Learning Working Group (authored by Microsoft and Google) that enables front-end web applications to expose callable tools, prompts, and resources directly to AI browser agents via document.modelContext.
Proposed by Microsoft, Google, and the W3C WebML Working Group. Expose your web application's native functionality via document.modelContext so AI agents can actuate your site with zero scraping.
Browser AI Agent Invocation
// Agent discovers & calls in-page tool
const tools = await document.modelContext.getTools();
const tool = tools.find(t => t.name === "filter-templates");
await document.modelContext.executeTool(tool, {
"description": "Spring flyer with date"
});Webpage Native Handler
// Page executes directly in client UI
execute({ description }) {
filterTemplatesInUI(description);
return {
content: [{
type: "text",
text: "Updated canvas with 3 matching templates"
}]
};
}The W3C Architecture
WebMCP bridges the gap between server-side Model Context Protocol and the living DOM.
Standardized web platform interface proposed under the W3C WebML Working Group. Register, manage, and unregister tools imperatively or declaratively.
Unlike backend MCP servers that bypass the browser, WebMCP executes in the user's tab, keeping active application state and visual UI in perfect sync.
Enforces browser security boundaries. Same-origin documents share tools automatically; cross-origin frames require explicit allow="tools" Permissions Policy.
Agent Ecosystem
When your website implements document.modelContext, AI browser agents call your actions directly instead of simulating mouse clicks or scraping HTML.
Dassi is built from the ground up with a native tool.webmcp discovery engine. It automatically detects declared tools and invokes them in under 5ms with zero scraping.
Connect web pages directly to Claude Desktop and local LLM clients using standard Model Context Protocol servers and client transports.
Authored under the W3C WebML Working Group by Google and Microsoft engineers, establishing an open cross-browser specification for browser-mediated AI actions.
Don't know where to start? Ask Dassi to inspect your web application. Dassi analyzes your UI elements, forms, and client state, and automatically drafts typed document.modelContext.registerTool() definitions ready to copy into your codebase.
Implementation Standard
Implement the official W3C specification in your front-end codebase today.
// W3C WebMCP Standard Tool Registration
const controller = new AbortController();
await document.modelContext.registerTool({
name: "search-products",
description: "Search in-stock products by query and price filters",
inputSchema: {
type: "object",
properties: {
query: { type: "string", description: "Search keyword" },
maxPrice: { type: "number", description: "Optional max price in USD" }
},
required: ["query"]
},
async execute({ query, maxPrice }) {
// Reuses existing client-side logic & updates page UI
const results = await app.catalog.search({ query, maxPrice });
return {
content: [{ type: "text", text: JSON.stringify(results) }]
};
}
}, { signal: controller.signal });
// To unregister dynamically on page/modal close:
// controller.abort();Live Diagnostic Utility
Verify if your webpage properly implements the W3C WebMCP specification (document.modelContext).
✓ Verified document.modelContext Standard
Target exports valid tool registration handlers compatible with AI browser agents.
Ecosystem
Let agents and users know your web application supports W3C WebMCP.
[](https://getwebmcp.dev)
Questions & Answers
WebMCP is an official web platform standard proposal authored by Microsoft and Google within the W3C Web Machine Learning Working Group (webmachinelearning/webmcp). It defines document.modelContext to let web pages expose client-side tools directly to AI browser agents.
Server MCP connects AI models to backend databases and local filesystems. WebMCP adapts this concept natively for browser tabs, allowing front-end applications to expose actions without writing a separate server daemon or replicating user authentication.
Include the lightweight universal polyfill shown in our Quickstart. It initializes document.modelContext and dispatches standard discovery events so AI browser agents (like Dassi AI and Chrome AI) can immediately discover your tools today.