WebMCP
A proposed web standard that lets sites expose structured tools to AI agents, so agents invoke defined actions instead of screen-scraping the page.
Today, when AI agents browse the web, they have to squint at websites, trying to guess how to click buttons and fill out forms. That screen-scraping process is incredibly fragile.
A new proposed web standard called WebMCP aims to change that. It allows a website to expose structured tools directly to AI agents. Instead of guessing, the agent receives a clear declaration of what the site can do, how to do it, and what inputs are required. This turns messy web-browsing into reliable, defined actions.
WebMCP is designed to be model-agnostic and is being developed as an open standard. Developers can implement it in two ways. The first is an imperative approach using JavaScript, where you register tools with a name, description, schema, and callback function. The second is a simpler, declarative approach using HTML. By adding basic attributes to your existing web forms, the browser automatically turns them into agent-ready tools, bringing the form into focus and waiting for user confirmation before running.
This standard builds on the Model Context Protocol, and it represents a foundational piece for the future agentic web.
WebMCP is a proposed web standard that lets a website expose structured tools directly to AI agents. Instead of an agent looking at your page and guessing what buttons and forms do, your site declares what it can do, how to do it, and what inputs it needs — turning fragile screen-scraping into reliable, defined actions.
There are two ways to implement it. The imperative API registers tools in JavaScript via navigator.modelContext.registerTool(), each with a name, description, JSON input schema and execute callback. The declarative API is simpler: add attributes like toolname and tooldescription to existing HTML forms and the browser turns them into agent-ready tools, bringing the form into focus and waiting for user confirmation on invocation.
Crucially it's model-agnostic and being developed as an open standard, not a Chrome-only or Gemini-only feature. It builds on the Model Context Protocol idea and is a foundational piece for the agentic web that AI agents will navigate.