What happened
Chrome M150 ships Lighthouse with a new Agentic Browsing category, announced June 22 as part of Google's agent-ready toolkit. The audit checks three things: whether every interactive element has a programmatic name in the accessibility tree, Cumulative Layout Shift, and whether the page registers WebMCP tools with valid schemas.
Each check maps to how agents actually operate. Agents read the accessibility tree as their primary model of your page — an unnamed button doesn't exist as far as they're concerned. Layout shift causes agent misclicks the same way it causes human ones. And WebMCP is the emerging spec for exposing site actions as callable tools an agent can invoke directly.
It arrives alongside Chrome DevTools for Agents, which can simulate agent interactions, run Lighthouse against an active tab, and debug via a screencast that exposes the accessibility tree. The tooling sits behind a Chrome flag and can be wired into coding assistants through mcp_config.json. Google is explicit that Agentic Browsing is informational and unbenchmarked — you get pass/fail statuses and warnings, not a score.
Why this matters
Google just published its working definition of an agent-ready site. Whatever you believe about agentic shopping timelines, this tells you exactly what Chrome's own tooling will measure — and informational audits have a habit of growing teeth later. If agents become a real purchase channel, the stores failing these checks are the ones agents abandon mid-task, silently, with no analytics event to mourn.
The accessibility check is the freebie. Fixing unnamed buttons and unlabeled controls helps screen-reader users today, trims accessibility legal exposure, and now makes your store operable by agents — three returns on one fix. Most ecommerce themes fail in the same spots: icon-only buttons, custom variant dropdowns, cart drawers rendered off-screen.
WebMCP is the speculative end of the audit. Registering tools for a spec this young probably isn't worth build hours for most stores yet. But the fact that Lighthouse already checks for it tells you where Chrome thinks this is heading.
What to do about it
Run the audit on your three money pages
In Chrome 150 or later, run Lighthouse with the Agentic Browsing category against your homepage, a top category page, and a best-selling product page. Save the results as your baseline — the point of an informational audit is knowing your gaps before anyone scores them.
Fix naming failures first
Every flagged element without a programmatic name is invisible to agents and screen readers alike. Start with add-to-cart buttons, variant pickers, and navigation controls — the elements an agent must find to complete a purchase on your store.
Reframe CLS as misclick risk
You've watched Cumulative Layout Shift for Core Web Vitals; this audit turns it into an agent-accuracy problem too. Recheck templates where late-loading banners, apps, or promo bars shove buttons around after first paint.
Put WebMCP on the watch list, not the roadmap
Note that the check exists and skip the build for now, unless you already run a custom app pipeline. Revisit when the spec and the audit both mature — early movers here are betting build hours on an unfinished standard.