WooCommerce for Claude – an experiment from Radical Speed Month

Automattic is currently running a structured experiment called Radical Speed Month. This post shares one of those experiments. It is not a feature announcement, not a roadmap commitment, and not a preview of anything Woo has decided to ship. It’s something we built, want to learn from, and would love your help testing.

What is it?

WooCommerce core already has MCP support in developer preview meaning AI clients can connect to a store and run basic product and order operations.

Our experiment, called WooCommerce for Claude, asks a different question:

• If AI clients can already talk to a Woo store, what does the intelligence layer on top look like?
• Does using Claude Code or CoWork to ask those questions provide additional value ?

This is not “let Claude call REST endpoints,” but rather “give Claude enough structured knowledge of the store that the answers are actually useful to a merchant.“

In practice that meant building three things on top of core MCP as a start:

• Analytics skills: pre-aggregated answers to questions merchants actually ask. (“How did my store do this week?”, “What’s driving my sales?”, “Are my coupons working?”) Each one runs against WooCommerce’s analytics lookup tables, not raw wp_posts.
• A knowledge layer: store profile, catalogue schema, policies, enriched product data. Exposed as MCP resources so the model has context before it makes a single tool call.
• An AI-readiness scoring engine: four weighted factors (product completeness, schema coverage, content quality, policy completeness) that produce a 0–100 score and a prioritised list of fixes.

Everything ships through one endpoint at /wp-json/woocommerce-claude/mcp, registered as WordPress Abilities and picked up by the WordPress MCP adapter that’s vendored inside Woo. No separate process. No cron. No background sync. The plugin only does work when an MCP request arrives.

What it looks like in Claude

Connect it using Claude Desktop or via your terminal to Claude Code. Then ask things like:

• My last 7 days look soft. What changed? Break it down by product, category, and time of day.
• Did my last promo lift revenue, or just shift it from full-price?
• Walk me through my store like I’m a new investor – what’s the headline, what’s the risk, what’s the upside?
• Where’s the leakage right now – biggest discounts, biggest refunds, and oldest on-hold or failed orders worth chasing?
• How much of my revenue is from repeat buyers, and which products bring them back?
• Audit my catalogue’s AI-readiness – where am I losing the most points, and what would I fix first?

Setup for Claude Desktop is a one-click .mcpb bundle. For other manual setup there is a copy-pasteable JSON snippet generated for your store, with a freshly minted read-only WooCommerce REST API key already filled in.

Why we’re sharing it now

Two reasons.

One, we want extension developers to try it out and see how you could integrate. The plugin uses a provider pattern meaning any extension can register itself and feed structured knowledge into the layer the AI sees:

add_action( 'woocommerce_claude_register_providers', function( $registry ) {
$registry->register( new My_Extension_Provider() );
});

You can also add a custom AI-readiness scoring factor, or filter enriched product data on its way out. The point of putting it in front of you is to find out whether this shape – providers + abilities + a single MCP endpoint – is one extension authors would actually build against.

Secondly, we want to test the actual ‘querying’ side with merchants and agencies and find the rough edges. The failure modes we most want to hear about are merchant-scope ones (does the AI suggest things the merchant can’t action?), privacy ones (does aggregated data ever leak something that should have been collapsed?), and performance ones on catalogues larger than our seeded demo store.

What we’re not claiming

• This is not a Woo feature. Nothing here is in core, in a release plan, or on an official roadmap. It’s a separate plugin, built under Radical Speed Month, and shared early. Some RSM projects will become permanent products later; many won’t. We don’t yet know which bucket this one falls into, and that’s part of what testing tells us.
• Not a finished product. Expect rough edges. Please open issues.

Try it out

• Repo: github.com/woocommerce/woocommerce-claude
• Install: clone, composer install, activate. Or grab the latest pre-release zip.
• Local dev: npx @wordpress/env start boots WP + Woo + the plugin on :8888. There’s a deterministic seed script that generates 24 months and 5,000 orders of realistic data so analytics responses have something to chew on.
• File issues in the repo: anything that surprises you, anywhere the AI gives a confidently wrong answer, anywhere the extension surface gets in your way.

If this experiment goes somewhere, your testing is what shapes it. If it doesn’t, your testing is what tells us why. Either outcome is the point of running RSM in the open.