Are you down with WebMCP? Google is.
WebMCP is a new Google protocol, now in early preview, designed to support AI agent interactions with websites. The goal is to help sites “talk to” agents by signaling how and where they should interact, making workflows smoother and enabling more complex agent-to-website tasks.
WebMCP proposes two new APIs that allow browser agents to take action on behalf of users: a Declarative API, which performs standard actions defined directly in HTML forms, and an Imperative API, which handles more complex, dynamic interactions that require JavaScript execution.
According to Google, the APIs “serve as a bridge, making your website ‘agent-ready’ and enabling more reliable and performant agent workflows compared to raw DOM actuation.”
Why This Matters:
AI agents have massive theoretical upside — some data suggest they could automate or assist with roughly half of common online work tasks — but that promise falls apart if the agent experience is slow, inaccurate, or clunky. If agents submit the wrong form field, misinterpret page elements, or stall mid-checkout, user trust erodes quickly. And without trust, adoption stalls.
That’s especially true in high-stakes, multi-step workflows. Think opening and populating customer support tickets, browsing and purchasing products on an ecommerce site, or searching for and booking flights. These tasks require precision and reliable execution. Speed and accuracy are prerequisites for scale.
Today, many agents rely on raw DOM manipulation, essentially guessing how to interact with page elements. That approach can work, but it’s brittle and prone to breaking when a site changes.
With WebMCP, Google is attempting to create more formal connective tissue between agents and websites — structured signals that tell agents how to act, rather than forcing them to guess. If successful, that could make agent interactions more stable, predictable, and scalable.
Experts React:
The adtech and developer community seem intrigued by WebMCP:
Our Take:
I know some might have “standard fatigue.” (From an acronym perspective, the alphabet has never been this employed.) But it’s encouraging to see progress on the technical plumbing behind agentic systems. None of this works if websites and agents aren’t speaking the same language under the hood.
