WebMCP is an early browser API proposal from the Edge and Google teams that lets web developers explicitly define which actions (called 'tools') AI agents can perform on their web pages. Using the Model Context Protocol (MCP), the browser acts as an MCP server, exposing developer-registered JavaScript functions to AI agents. This gives developers control over how agents interact with their pages rather than relying on DOM scraping or screenshots. A demo shows VSCode Copilot in agent mode connecting to Edge's MCP server to interact with a TODO list app. The author is seeking early feedback on the proposal.

6m read timeFrom patrickbrosset.com
Post cover image
Table of contents
AI agents and the webWebMCPExampleDemoFeedback

Sort: