Skip to main content
WebMCP is a Community Group proposal that lets a web application publish structured actions as tools for AI agents. The site defines the actions and their inputs. The agent uses that contract instead of inferring behavior from screenshots, DOM structure, or click sequences. The browser is a useful home for these tools because the page already has the application’s state, authorization checks, and visible interface. A tool can reuse that context without moving browser-only logic into a separate service. The application still validates inputs, enforces permissions, and asks for confirmation when the consequence warrants it.

​
Learn about the proposal

The Community Group draft owns the proposed JavaScript API. Chrome maintains the current developer overview, user journey guide, and WebMCP and MCP comparison. The WebMCP API sources and declarative API sources route implementation questions to their upstream owners. WebMCP resources and status collects status, security, inspection, and testing resources.

​
Where MCP-B fits

MCP-B provides packages around the proposal, including a polyfill, React bindings, transports, iframe bridges, and a local relay. Bridges can make page tools available to MCP clients, but the browser document remains the execution environment. MCP-B does not define the WebMCP proposal. WebMCP and MCP-B extensions explains which behavior belongs to this project. Transports and bridges explains how those packages cross browser boundaries.