WebMCP

WebMCP launched in early preview in February 2026 inside Chrome 145. The core problem it addresses: AI agents that operate in browsers currently have to simulate human behavior, taking screenshots, parsing the DOM (a web page's underlying structure), guessing where buttons are, and clicking through menus. That approach is slow, fragile, and expensive in tokens. WebMCP flips this by letting website developers explicitly declare what actions their site exposes, with typed inputs and outputs, so an agent can call them directly like API functions.

There are two flavors. The Declarative API works with existing HTML forms: add a tool name and description to a form and it becomes agent-callable without JavaScript changes. The Imperative API lets developers register richer tools via JavaScript, defining schemas (structured descriptions of inputs and outputs) similar to what you'd send to the OpenAI or Anthropic APIs, but running client-side in the browser. The spec was developed with Microsoft and is moving through the W3C standardization process, with both Google and Microsoft shipping initial code.

For builders, WebMCP matters in two directions. If you're building AI agents that interact with third-party websites, a WebMCP-enabled site can slash your token costs and reliability failures since one structured tool call replaces dozens of screenshot-and-click loops. If you're building a web product, implementing WebMCP makes your site agent-ready: you control exactly which actions agents can take and what data they get back, creating a cleaner audit trail than scraping traffic. Adoption is still early as of mid-2026, but Cloudflare already added WebMCP support to its Browser Run product.