WebMCP turns page actions into explicit tool contracts.

Chrome describes WebMCP as a proposed web standard for exposing structured tools to AI agents. A page can register imperative JavaScript tools or annotate HTML forms declaratively. The browser presents tool names, descriptions and JSON schemas within the current page context.

This differs from pure actuation, where an agent interprets each visual or DOM step and simulates input. It also differs from a remote MCP server because the tool is discovered and executed in a visible browser context tied to the current origin and page state.

As of this review, Chrome documents WebMCP as proposed and experimental, with origin-trial requirements. Client support and APIs can change. Its absence should not automatically fail a website that offers a reliable accessible UI or safe API path.

Use WebMCP where an explicit contract materially reduces ambiguity, such as complex filters, support routing or date selection. Do not expose a tool merely to increase an inventory count.

A useful tool has a precise name, purpose, input schema, structured result and predictable error model. The contract should distinguish read-only operations from actions that create, modify or delete state.

Chrome documents origin isolation and a tools Permissions Policy for WebMCP. Those platform gates are only the start. The application must still enforce authentication, authorization, tenancy, rate limits and business constraints below the browser layer.

Use deterministic tests for schema validation, business rules and final states. Add model evals for whether agents choose the correct tool, supply acceptable parameters and interpret the result. Record failures separately for the site, tool, model, policy and infrastructure.