Google Officially Announces WebMCP: Everything from the Feb 10, 2026 Launch

On February 10, 2026, Google made WebMCP official. Here is the announcement recap, the rollout timeline, and the three things every web developer should do this week.

On the morning of February 10, 2026, Google's Chrome team published the post every agentic-web watcher had been waiting for: WebMCP is officially an Early Preview Program, shipping in Chrome 146 Canary behind a flag, with a full stable rollout planned for mid-2026. The announcement, written by Chrome developer advocate André Cipriani Bandarra, marked the end of two years of quiet prototyping and the start of the migration wave.

What was announced

• WebMCP is now an official Chrome proposal, developed in the open at webmachinelearning.github.io/webmcp, with cross-vendor W3C collaboration.
• Chrome 146 Canary ships native support behind the chrome://flags#webmcp flag. Once enabled, navigator.modelContext is a real object, not a polyfill.
• The Early Preview Program (EPP) is open to all developers. Sign up via the EPP landing page; Google triages feedback weekly.
• Stable rollout targeted for mid-2026, pending spec finalization and at least two independent implementations (Firefox and Safari have expressed interest).
• Reference implementation and docs are published alongside the announcement, plus a test page at Chrome's developer preview site.

Why the timing matters

WebMCP has been discussed in W3C circles since 2024, but the protocol needed a browser willing to ship a native implementation. Google doing so in Canary forces the issue: Firefox and Safari will either align on the spec or risk becoming the browsers that AI agents can't use. That's a fast-moving competitive dynamic.

For site owners, it means the window where WebMCP is "optional" is closing. Early EPP adopters get an entire product cycle to refine their implementation before their competitors notice.

The three tool surfaces reaffirmed

The announcement formalizes the three API surfaces most of us had been using from the draft spec:

1. Declarative API — HTML form attributes (toolname, tooldescription). No JavaScript needed.
2. Imperative API — navigator.modelContext.registerTool() plus friends.
3. .well-known/webmcp manifest — static JSON for out-of-band discovery by crawlers and agents.

These three were present in the draft; the launch doesn't change them. What did change: human-in-the-loop is now a normative part of the spec, not a best practice — requestUserInteraction() must be supported by conforming implementations.

What developers should do this week

1. Install Chrome Canary. Enable the WebMCP flag. Open DevTools → Application → WebMCP and confirm the panel is there.
2. Read the spec. webmachinelearning.github.io/webmcp — 40 pages, readable in an hour.
3. Audit your own site. Run our free WebMCP readiness check — it takes 15 seconds and gives you a scored report.

What this announcement doesn't say

Three things the post did not commit to, which are worth watching in the next month:

• Payment flows. The announcement references human-in-the-loop for payments but defers actual payment-token integration to "a future revision." Stripe and Adyen have both hinted at first-party support; no dates yet.
• Mobile. Chrome Android gets WebMCP in Canary, but deeper OS integration (share-sheet, cross-app) is unlisted.
• Enterprise policy controls. IT admins need a policy to disable or audit WebMCP across a fleet. Google's enterprise team confirmed it's "in progress."

How to follow the rollout

The authoritative source is the Chrome WebMCP EPP blog. For practical deep-dives, keep an eye on our API reference, the full implementation guide, and the readiness checklist. We update all of them whenever the spec moves.

Frequently Asked Questions