Google introduces WebMCP with Chrome DevTools for agents and Modern Web Guidance

• Google framed the drop as a package of browser-facing agent tooling, with Google's announcement thread tying together native Android generation in AI Studio, Modern Web Guidance, WebMCP, and DevTools support.
• Google's WebMCP post described WebMCP as a proposed open web standard that lets websites expose functions to coding agents more accurately than brittle page scraping.
• Google's DevTools for agents post said Chrome DevTools can now be driven by agents to test code, emulate users, and catch bugs before shipping.
• Google's Modern Web Guidance post positioned the guidance as a blueprint of web-development skills for agents, which is a different bet from just giving models bigger context windows.
• A keynote checklist spotted by Jeremy Howard's photo adds two concrete hooks Google did not spell out in the thread, WebMCP tool validation and llms.txt verification inside DevTools.

You can read Google's WebMCP post as a standards move, the DevTools announcement as a browser automation move, and Jeremy Howard's keynote photo as the hint that Google wants validation features around both. The HTML-in-Canvas launch post also landed in the same thread, which makes the package feel less like a single API launch and more like Google trying to wire agents directly into the modern web stack.

Google's core claim is simple: websites need a cleaner way to expose capabilities to agents than DOM poking and handwritten browser scripts.

According to Google's WebMCP post, the proposal is meant to bridge websites and coding agents so they can call site functions more accurately, precisely, and reliably. That makes WebMCP the most ambitious part of the bundle, because it tries to standardize the website side of agent behavior, not just the model side.

Google paired the protocol pitch with a debugging surface that already exists inside Chrome.

Google's DevTools post says agents can use Chrome DevTools to verify code, emulate users, and debug in real time. For engineering teams, the interesting part is not the browser control itself, it is that Google is packaging verification and debugging as first-class agent tasks rather than leaving them to custom harnesses.

Modern Web Guidance is the quieter piece, but it fills in the missing part of the stack: what a competent web-building agent should know.

Google's description calls it an early preview and says it gives agents a blueprint for modern web development across many use cases. In practice, that reads like Google trying to turn accumulated frontend best practices into a reusable skills layer for agents.

The same thread also shipped a separate browser primitive, and a keynote slide exposed extra validation work around the agent stack.

Google's HTML-in-Canvas post says the origin trial lets developers place real DOM elements inside canvas so immersive interfaces stay searchable, accessible, translatable, and connected to browser features. Separately, Jeremy Howard's keynote photo showed a checklist with "Validate WebMCP tools," "Verify llms.txt files," and "Check accessibility," which is the clearest evidence here that Google's agent tooling plan reaches beyond generation into conformance checks.