WebMCP is an experimental Chrome feature that turns any web page into an MCP (Model Context Protocol) server, enabling AI agents to interact with web pages in a structured way. Backed by Google and Microsoft, it uses a Chrome flag and extension to expose page actions as MCP tools. The setup involves joining Google's Early Preview Program, enabling a Chrome flag, and installing an inspector extension. Two APIs are proposed: a Declarative API for standard HTML actions and an Imperative API for JavaScript-driven interactions. A live demo with a mock flight search shows agents querying page data and receiving structured JSON responses. Importantly, the human user remains in the loop — WebMCP supports both agent-driven and human-assisted browsing sessions.
