So Google and Microsoft quietly shipped something in Chrome 146 Canary that I think most people are sleeping on. It’s called [WebMCP](https://developer.chrome.com/blog/webmcp-epp) — short for Web Model Context Protocol — and it basically lets websites expose structured, callable tools directly to AI agents running in the browser. No more scraping. No more pixel-level screenshot parsing. Just clean function calls.

Here’s why this matters. Right now, if you want an AI agent to book a flight or search a product catalog on a website, the agent has to do this awkward dance of clicking through dropdowns, scrolling paginated results, and screenshotting every page. It’s slow, expensive, and brittle. WebMCP flips that on its head. A site can register something like a `searchProducts` tool, and the agent just calls it directly and gets structured JSON back. One call instead of dozens of clumsy browser interactions. Early benchmarks are showing [roughly 67% less computational overhead](https://venturebeat.com/infrastructure/google-chrome-ships-webmcp-in-early-preview-turning-every-website-into-a) compared to the old visual-agent approach, which is a pretty staggering number.

The spec offers two flavors. The Declarative API is dead simple — you basically annotate your existing HTML forms with tool names and descriptions, and they become callable. If your site already has well-structured forms, you’re most of the way there. The Imperative API is the heavier-duty option, letting you define richer tool schemas in JavaScript for more dynamic interactions. Both run client-side through a `navigator.modelContext` API. The whole thing is being incubated through the [W3C Web Machine Learning Community Group](https://github.com/webmachinelearning/webmcp), with Google and Microsoft engineers co-authoring the spec.

What really caught my attention is the pace of adoption buzz. The [developer preview landed on February 10th](https://developer.chrome.com/blog/webmcp-epp), and by the 13th [VentureBeat](https://venturebeat.com/infrastructure/google-chrome-ships-webmcp-in-early-preview-turning-every-website-into-a) and a bunch of other outlets were all over it. There’s already an active [Hacker News thread](https://news.ycombinator.com/item?id=46997184) with people calling it “the USB-C moment for AI agents,” which honestly isn’t that far off. When the two biggest browser vendors are co-shipping a standard like this, it tends to actually stick. People are already speculating about formal announcements at Google I/O later this year.

If you build anything on the web and you’re even remotely thinking about AI agent compatibility, this is worth paying attention to right now. The [examples repo](https://github.com/WebMCP-org/examples) has some solid starting points. We’re watching the web grow a whole new interface layer in real time, and it’s moving fast.