Google proposes WebMCP open standard

Google has published a proposal called WebMCP that would give websites a standardized way to expose actions to AI agents running in browsers, according to Chrome developer documentation released on May 18. The proposal is framed as an open web standard and is documented both on Chrome for Developers and in a public GitHub repository maintained through the Web Machine Learning Community Group. (developer.chrome.com) Google says the goal is to let websites declare what an agent can do, rather than forcing agents to infer actions from buttons, fields and page layouts. (developer.chrome.com) The materials describe WebMCP as a way to improve the “performance and reliability” of agent actions on the web. ### What is Google actually proposing with WebMCP? WebMCP is a proposed web standard that lets a site expose “structured tools for AI agents,” Google said in its Chrome documentation published May 18. The company said websites can use JavaScript and annotated HTML form elements so that agents know “exactly how to interact with page features.” The GitHub explainer says web applications using WebMCP can be treated like Model Context Protocol servers implemented in client-side script rather than on a backend. The repository lists contributors from Microsoft and Google, including David Bokan, Khushal Sagar, Brandon Walderman and Leo Lee, and says the project was first published on Aug. 13, 2025. (developer.chrome.com) ### How is this different from the existing MCP standard? Google said on March 11 that WebMCP “is not an extension or a replacement of MCP.” In that post, Google described MCP as a backend-facing protocol for connecting agents to external systems, while WebMCP was described as a frontend browser standard for interacting with a site’s user interface. (developer.chrome.com) The same post said both approaches share the goal of structured tool discovery and predictable execution, but differ in where the functionality lives. MCP is available across platforms and often uses JSON-RPC, Google said, while WebMCP is limited to the browser and is implemented with JavaScript or HTML attributes. (github.com) ### What does the API look like on a website? Google’s early-preview post on Feb. 10 said WebMCP proposes two APIs: a declarative API for standard actions defined in HTML forms, and an imperative API for more complex interactions that require JavaScript execution. Google said those interfaces are intended to make a website “agent-ready.” (developer.chrome.com) The imperative API documentation published May 18 shows developers registering tools through `navigator.modelContext.registerTool`. The examples include a `toggle_layer` tool and a `get_order_status` tool, each with a name, description, input schema and execute function. Google said this interface can support actions including form input, site navigation and state management. (developer.chrome.com) ### Does Google say this is more reliable than screen scraping? Google said WebMCP is intended to reduce guesswork by letting the website declare an element’s purpose directly. In its Chrome documentation, the company contrasted that with “actuation,” which it defined as an agent simulating clicks and text input as if it were a human user. (developer.chrome.com) The documentation says WebMCP supports discovery, JSON schemas and shared state so the agent can understand available resources and expected inputs and outputs in real time. Google said that can improve accuracy for tasks such as support flows, ecommerce and travel booking. (developer.chrome.com) ### Are Apple and Mozilla signed up? Public Google materials reviewed for this story do not include a commitment from Apple or Mozilla to ship WebMCP. The Chrome documentation says Google’s goal is to build APIs “that any browser with agentic capabilities can implement and benefit from,” and points developers to the public GitHub process. (developer.chrome.com) The public GitHub repository does show cross-company participation, including Microsoft and Google contributors, but the sources reviewed here did not verify a formal adoption plan by Apple or Mozilla. ### What happens next for developers? (developer.chrome.com) Google said on Feb. 10 that WebMCP was available to early preview program participants for prototyping. The May 18 documentation also points developers to Chrome Status, a developer trial and GitHub discussion channels, and says the API is under active discussion and “subject to change.” The next concrete step is in Google’s own materials: developers can join the early preview program, review the explainer on GitHub, and test the imperative and declarative interfaces in Chrome’s current trial documentation. (developer.chrome.com 1) (developer.chrome.com 2) (github.com)