Announcement · April 16, 2026

The AI your next customer asks isn't going to guess what your site does.

In a few months, a shopper will tell ChatGPT or Claude "book me a plumber" or "buy me a coffee grinder under $200" and expect the agent to finish the job. The agent will cite exactly one business. Maybe two. That business will be the one whose site told the agent what tools it offers — before the agent ever loaded the page.

The web standard for that conversation is called WebMCP. The W3C published a Draft Community Group Report on February 10, 2026. Chrome 146 Canary ships an early preview behind a flag. And it has one gap your site is about to fall through.

What AI agents can't see, they can't recommend.

WebMCP lets your site register tools the browser can hand to an AI: navigator.modelContext.provideContext({ tools: [...] }). Add to cart. Check availability. Place order. The W3C draft is good. The browser preview is real. But there's a catch every early adopter is already hitting:

AI agents can't discover your tools before they load your page.

External: your site registers placeOrder at runtime. An agent planning a purchase has to execute your JavaScript just to find out that tool exists. Too late to plan against. Too late to verify it's safe to call.
Internal: you've been shipping for humans for a decade. Now the top of your funnel is an AI. If your stack can't answer the agent, a competitor's stack will. You feel the shift and you don't yet have a playbook.
Philosophical: the web shouldn't decide who exists based on which proprietary demos invited them in. Agents deserve a signed, verifiable inventory of what a site can do — served openly, the way HTML and HTTPS are served openly.

We've been in the agent-readiness trenches with you.

Speakable has been scoring and certifying businesses for AI commerce since before ChatGPT Instant Checkout shipped. We watch every protocol the agent economy runs on — UCP, A2A, Google AP2, Mastercard Agent Pay, Visa TAP, Web Bot Auth, Stripe ACP. Our Research Agent checks the W3C WebMCP repo daily. Today that repo has 79 open issues and the latest one was opened this morning.

You don't need to track all of that. You need one thing to work. We built it.

Three steps from invisible to cited.

Step 01 · Scan

Run a free Speakable Score on your URL. In under two minutes we score your site 0–100 across Discoverability, Verification, Transactability, and Certification — including a 0–10 WebMCP sub-score and a 0–7 transactional-surface score per the spec we just published.
Step 02 · Fix

Generate a signed manifest with one API call. Our Remediation Agent builds a ready-to-deploy /.well-known/webmcp.json signed with the same Ed25519 key chain as your Speakable Trust Certificate, plus a runtime JS snippet with provideContext() and registerTool() calls wired to your backend, plus a copy-paste install doc for Shopify, WooCommerce, or any origin.
Step 03 · Monitor

Sleep through the drift. Our cron re-verifies every certified manifest every 24 hours. Signature regresses, manifest disappears, new transact-class tool appears without user-interaction gating — you get an urgent alert before any AI agent calls a tool it shouldn't.

Our commitments

✓ Open-spec — no proprietary extensions
✓ Forward-compat with whatever the W3C ships
✓ Ed25519 signing, same key chain as your trust cert
✓ Drift-monitored daily — you never silently fall out of compliance
✓ No vendor lock-in — manifest works without us
✓ Free scan, no card required

Run your free Speakable Score →

Without a WebMCP manifest

Agents skip you for competitors who declared their tools
Every transact-class tool you ship runtime-only is unverifiable
You discover drift on Twitter, not in your dashboard
Certified tier stays out of reach

With one

Pre-flight discoverability — agents plan before they visit
Signed tool inventory, cryptographically tied to your domain
Daily drift monitoring with urgent alerts on tamper
Direct path to the Certified tier

We're the first ones certified by this standard.

Speakable's own platform serves its signed WebMCP manifest at /.well-known/webmcp.json — four read-class tools, signed with the same Ed25519 key as our trust certificate. We publish the spec we ask merchants to implement. We pass the scan we ask merchants to pass. If our manifest drifts from our runtime, our own cron opens an alert on our own dashboard — and you can audit the whole thing.

Transparency artifacts:

Pick your path.

The agent layer is forming now, with or without you. You can be the business AI agents cite when a shopper asks, or you can be the one they skip because they couldn't verify what you do.

Run your free Speakable Score → Read the spec first

Two-minute scan · No credit card · Cited sources for every claim