WebMCP urges sites to serve agents

Search Engine Land published an article on May 15 urging businesses to prepare for WebMCP, a proposed web standard designed to let AI agents discover and use site functions directly rather than infer them from page layouts. Dave Davies wrote that companies should treat the technology as a potential new discovery layer as chatbots and browser-based agents take on more navigation and task completion. The argument lands as Google Chrome’s developer documentation and a W3C Community Group draft show the idea has moved beyond a loose concept into early technical implementation. The W3C Community Group draft dated April 23, 2026 describes WebMCP as a JavaScript interface that allows web applications to expose functionality as tools with natural-language descriptions and structured schemas. Chrome Developers said in February that the goal is to let AI agents perform actions on sites with greater speed, reliability and precision. Search Engine Land framed that shift in practical terms for marketers and site owners: if agents become intermediaries, sites may need to expose actions in machine-readable ways, not just publish pages for human visitors. (searchengineland.com) ### What is WebMCP supposed to change on a website? The April 23 draft says WebMCP would let developers expose site functions as tools that agents can invoke, using structured definitions for inputs and outputs. The specification says pages using WebMCP can be thought of as MCP servers implemented in client-side script rather than only on the backend. Chrome Developers and Search Engine Land both describe the current problem as one of guesswork. (webmachinelearning.github.io) Without a structured interface, an agent trying to book a flight, check out a cart or submit a form has to inspect the page, identify fields and buttons, and infer what to do next. With WebMCP, Chrome’s documentation says, the site can declare those actions directly. ### How is that different from scraping a page or reading HTML? Chrome’s March 11 guidance drew a line between MCP and WebMCP by saying MCP is for backend systems, while WebMCP is for frontend interactions on open pages. The same guidance says WebMCP tools are ephemeral: they exist only while the page is open, and the agent loses access when the user leaves the site or closes the tab. (searchengineland.com) Chrome’s March 4 explainer said the change is from agents searching for visible elements to agents calling named functions with defined parameters. In its example, a flight-booking tool would accept structured fields such as origin, destination, date and passenger count, then return a structured result such as a confirmation number and price. (developer.chrome.com) ### What would a company actually have to publish? Chrome’s developer materials say WebMCP can work through JavaScript APIs and HTML form annotations. Lighthouse documentation published in May says registered WebMCP tools are the capabilities a site exposes to agents, while a separate schema-validity audit checks whether required metadata and field names are present. (searchengineland.com) Chrome’s Lighthouse guidance also points to concrete implementation details. A forms audit says developers can add declarative metadata such as `toolname` and `tooldescription` to forms, and another scoring page says sites can improve “agentic readiness” by adopting WebMCP and maintaining semantic HTML and proper ARIA labeling. ### Why are marketers and publishers paying attention now? (searchengineland.com) Search Engine Land’s May 15 article argues that businesses should prepare before the standard is mature because discovery behavior is already shifting toward assistants and agents. That is an argument from Davies, not a formal standards body position, but it reflects a broader concern in search and publishing that traffic and transactions may increasingly pass through machine intermediaries rather than direct page visits. (developer.chrome.com) Chrome’s own materials show the concept is still experimental. The Chrome Developers blog said in February that WebMCP was available through an early preview program, and the Chrome AI documentation says it is available for prototyping to early preview participants. ### Where can developers test this now? Chrome Developers said WebMCP is available in an early preview program and developer materials tie current testing to Chrome 146 and newer experimental tooling. (searchengineland.com) Search Engine Land’s March 4 article said Chrome 146 introduced an early preview behind a flag, while Chrome’s DevTools update last week said experimental tools can now list and execute WebMCP tools exposed by pages and Lighthouse now includes an “Agentic Browsing” category. (developer.chrome.com) The next concrete step is in Chrome’s documentation. Google’s developer pages now point participants to the early preview program, Lighthouse audits for registered tools and schema validity, and setup guidance for testing WebMCP implementations in Chrome. (developer.chrome.com) (searchengineland.com)