What is WebMCP? Exposing website tools to browser agents
WebMCP tool registration and form interactions, distinct from MCP servers, browser automation and account permissions, with community-specification and preview limits.
Understand the concept
WebMCP is a browser-interface proposal for websites to expose structured tools to AI agents. Sites describe actions and inputs so supporting agents can call them without finding every control. Its related name does not mean existing MCP servers run unchanged in webpages, and it is not a model. Check specifications and implementations instead of assuming universal support.
Define action names, arguments and execution logic
A hypothetical order-query tool accepts an identifier and executes existing site logic. The model selects the tool while browser and site execute and return results. No real orders were queried. Structured arguments reduce dependence on button position without preventing wrong objects or fields.
Definitions need clear purposes, inputs and usable results. Vague names do not ensure reliable operation. Existing server authorization remains required for protected actions; registration does not create account access.
Forms and JavaScript registration suit different actions
Chrome’s preview distinguishes declarative form actions from imperative JavaScript for dynamic interactions. The community specification describes JavaScript registration, input Schema and execution. Interfaces may evolve.
Sites can retain human interfaces while supplying explicit actions to supporting consumers. Check browser versions and enablement rather than assume adding code makes every assistant discover or invoke it.
Deployment and connection layers differ from MCP
MCP connects clients and servers with tools and information, whereas WebMCP exposes webpage actions to browser agents. Deployment and access differ; renaming does not establish interoperability. Existing servers do not automatically become page tools, and page tools need not be remote MCP servers.
The current community specification is neither a W3C Standard nor on its Standards Track. Chrome’s early preview is distinct from stable cross-browser support. These conditions are not permanent compatibility guarantees; verify actual specifications and implementations before adoption.
Callable tools do not establish user authorization
Structured tools can reduce some uncertainty in locating interface elements, but cannot eliminate model misselection, argument-validation failures or prompt injection. Website text and returned data retain source roles instead of becoming user instructions. Writes require authorization and confirmation semantics rather than bypassing ordinary limits.
Compare with screenshot interaction according to available tools and real support. Confirm business outcomes rather than accept a success string alone. For argument failures, inspect the actual Schema and objects. A WebMCP label does not establish error-free automation.
Checked sources
These sources support the concepts and methods described here. Examples and test scope are labeled separately; check the version of the tools you use.