There is a recurring gap in how AI agents interact with web services. An agent that needs to query a database, file a form, or complete a transaction has two options: it can use a well-designed API if one exists, or it can scrape, click through, and hope the interface does not change. The first path is reliable and fast. The second path is brittle and slow. Most websites only offer the second path, because they were built for human eyes.
Google's announcement of WebMCP at I/O 2026 is an attempt to change this at the platform level. The Chrome for Developers blog describes it as "a proposed open web standard allowing developers to expose structured tools — like JavaScript functions and HTML forms — so browser-based AI agents can execute complex tasks with greater speed, reliability, and precision." By defining these tools explicitly, developers instruct agents exactly how to interact with their site: which functions to call, which forms to fill, which endpoints to use. The experimental origin trial begins in Chrome 149.
The distinction from MCP — the Model Context Protocol standard that has spread rapidly since Anthropic introduced it — is worth being precise about. MCP connects agents to external services via a server that an operator runs. WebMCP is native to the web page itself: developers publish tool definitions as part of their site, and any agent running in a compliant browser can discover and use them. The server is the website. There is no separate integration layer to build.
This is a web platform play. Google is trying to establish the open web as the preferred substrate for agents at the same moment that proprietary app ecosystems and walled-garden APIs are competing for that position. The argument is the same one that made HTTP and HTML the dominant protocols for human interaction: open and composable beats closed and proprietary over time, because the open version attracts more contributors and spreads to more surfaces.
The Google Developer Blog's I/O roundup frames the broader vision: WebMCP is one component of what Google is calling the "agentic web era," alongside Chrome DevTools for agents (which we covered last week), Modern Web Guidance for building agent-accessible experiences, and Gemini 3.5 Flash, which Google positions explicitly as a model designed for agentic workflows.
For a developer running a web service today, the practical consideration is this: websites that do not implement WebMCP tooling will still be accessible to agents — via scraping, clicking, and parsing. But websites that do implement it will be discoverable and callable in a way that makes them first-class citizens in agentic workflows. The early adopter advantage is real, because agents optimized for WebMCP-compliant sites will prefer them over the scraping path, and users whose agents can complete tasks reliably on your site will return.
The political stakes are real even if the standard is early. MCP, which Anthropic published, has already spread far enough that it would be difficult to displace as the server-side integration layer. WebMCP is Google's move for the client side — the browser layer, which Google controls through Chrome. If the standard sticks, Google is not just a model provider. It is the infrastructure owner of the agentic web.
The origin trial is just Chrome 149. This is a proposal, not a fait accompli. Standards fail. What makes this one worth paying attention to is who is proposing it: not a startup, not a research lab, but the company whose browser 65% of web users worldwide are using. If Google ships WebMCP into stable Chrome, it is a standard whether the W3C adopts it or not.
For builders working in emerging markets — where the friction of maintaining separate API integrations for every service is compounded by thin engineering teams — the WebMCP model is appealing. A service defines its tooling once, and any agent can use it. In most emerging markets, including Mongolia, Chrome holds more than 60% market share; betting on Chrome's dominance for the agent layer is a reasonable call.
The web is not going away. But whether it becomes an active participant in agentic workflows, or a passive surface to be scraped, is a question that is being answered right now.
The short of it.
At Google I/O 2026, Google proposed WebMCP — a web standard that lets developers expose structured tools directly from their websites so AI agents can call them without scraping or fragile automation. The origin trial starts in Chrome 149. Unlike MCP, which requires a separate server, WebMCP is native to the page itself. This is a structural bet: Google wants the open web to be the substrate for the agentic era, not proprietary app ecosystems. Builders who define their tooling now will be callable by agents; those who do not will be scraped.