WebMCP aims to give websites a structured interface for AI agents. A search, availability check or form can be described as a tool, so the agent does not have to infer every interaction from the page layout.
As of September 30, 2026, it is a Draft Community Group Report from the W3C Web Machine Learning Community Group. The document explicitly is neither a W3C Standard nor on the W3C Standards Track. Google and Microsoft provide the editors. This is a concrete technical initiative, but it does not mean every browser or assistant already supports it.
What WebMCP provides#
WebMCP is a browser API. The current draft uses document.modelContext; older examples use navigator.modelContext. A website registers JavaScript functions with names, descriptions and input schemas. An execute callback runs the function.
Tools live in the open document. A file at /.well-known/webmcp or an agents.json does not replace this registration. WebMCP defines neither a minimum number of tools nor a general catalog of MCP resources and prompts.
A restaurant could offer a read-only menu search. A real estate business could filter listings by location and price. Booking also requires a working booking process and clear authorization for the action.
Draft, browser and agent are separate questions#
Chrome introduced WebMCP as an early preview. A browser version containing an experimental implementation does not mean it is enabled by default for everyone. An available API also does not mean ChatGPT, Claude or Gemini automatically call tools on every website they visit.
Before integrating it, check the browser, current API name and actual agent. Start with Chrome's early preview announcement and the WebMCP specification. Keeping the normal user interface as a fallback is sensible.
Plan a useful integration#
- Choose a real function, such as product search or an availability check.
- Clearly describe inputs, results and potential side effects.
- Register the tool only when the browser supports the required API.
- Check authentication and permissions in the application itself. A description or annotation does not replace that check.
- Test both invocation by a compatible agent and normal use without WebMCP.
The draft discusses malicious tool descriptions, harmful outputs and unintended actions. Orders, contact requests and bookings therefore need a clear confirmation step in their workflow.
What it means for visibility#
WebMCP can make a website easier to operate for agents that support it. That does not establish a Google ranking benefit, preferential inclusion in AI Overviews or guaranteed new customers. Fixed overhead or accuracy figures only apply to a particular measurement with a known method and scope.
For a small business, the useful question is whether an agent could safely carry out a recurring action with clear inputs. If so, a limited trial may help. If the site mainly provides information, clear content, usable interfaces and accurate structured data remain the foundation.
At StudioMeyer, machine-readable information and suitable interfaces are part of website planning. See our web design offering for the scope of a project. We match a WebMCP integration to the actual process and available browser and agent support.
