UCP Checker
Developer Reference

UCP Product Discovery

How AI shopping agents find and buy products on UCP-verified stores. The MCP tool call sequence, capability requirements, and a live list of 19,004+ stores supporting it right now.

In 60 seconds

What product discovery actually means for AI agents

Product discovery is how an AI shopping agent figures out what a store sells without scraping HTML or guessing. It's the difference between an agent that might find the right product and an agent that's guaranteed to.

On a UCP-verified store, the whole flow is structured. The agent fetches the store's /.well-known/ucp profile, sees which capabilities are declared (the important one here is catalog-search), and dispatches tool calls against the declared transport endpoint. Each call returns typed product objects the agent can reason about directly — no HTML parsing, no fragile selectors, no guessing at price formats.

Below is the live demo: a real AI agent running the full discovery → browse → checkout sequence against a real UCP-verified store. Every tool call is what your own agent would send.

● A real AI agent doing product discovery on a verified UCP store

The Sequence

The four steps that make up discovery

Every UCP product discovery flow boils down to the same four steps. The first reads the store's profile; the other three are MCP tool calls to the endpoint the store declares, using the tool names the UCP binding defines. All of it happens on the store's own endpoint: UCP Checker tells an agent which stores to try and never sits in the transaction.

  1. 1

    /.well-known/ucp

    Agent fetches the store's profile and reads it. This is the handshake: the agent sees which capabilities the store declares, which transports and endpoints it exposes, and which payment handlers it accepts. If catalog search (dev.ucp.shopping.catalog.search) isn't declared, discovery stops here. UCP Checker's check-domain and discover-store tools report the same thing, as observed.

    HTTP GET · JSON response
  2. 2

    search_catalog

    Agent calls search_catalog on the store's declared MCP endpoint, with the query nested under the operation ({catalog: {query: "wool sneakers", context: {address_country: "US"}}}) and its platform profile in meta["ucp-agent"]. The store returns a typed list of product objects — title, price, currency, variants, image URL — with no HTML parsing required. This is where "scraping vs shopping" stops being an analogy and becomes a measurable capability gap.

    MCP tool call · returns products[]
  3. 3

    create_cart

    Agent picks a variant and calls create_cart (and update_cart to change it). The store returns the cart and a checkout URL. At this point the agent has a real, persistable cart — not a simulation, not a preview.

    MCP tool call · returns cart + checkout URL
  4. 4

    create_checkout → complete_checkout

    Agent opens a checkout with create_checkout, pays through one of the store's declared payment handlers, and calls complete_checkout. The store returns the order. The agent has bought something — end to end, no human in the loop.

    MCP tool calls · returns order
Try It

Run product discovery yourself

Three tools, three roles. UCP Checker tells you whether a store supports discovery. UCP Playground runs a real AI agent session against the store so you can watch every tool call live. UCP Registry exposes a discovery API you can call from your own app. They're designed to be used together.

Validation

Check a store's discovery readiness

Paste any domain and UCP Checker fetches its profile, validates against the current spec, confirms it declares catalog-search, and returns the full capability breakdown. Free, no signup, runs in ~1 second.

Run Checker
Agent Sessions

Watch a real agent shop a store

UCP Playground runs a live AI agent session against any verified store — discover, search, cart, checkout. Watch every MCP tool call and response in real time. Zero code, just pick a store and press play.

Open Playground
Discovery API

Run discovery from your own app

UCP Registry exposes a live product discovery API — pass a store domain and a query, get typed product results back. The production endpoint for developers who want to actually do discovery at scale, not just test it.

Open Registry

Frequently Asked Questions

What is UCP product discovery?
The process by which an AI shopping agent finds products on a store without scraping HTML. The agent fetches the store's /.well-known/ucp profile, sees which capabilities are declared, and calls search_catalog on the MCP endpoint the store declares. The store returns typed product objects the agent can reason about directly — no parsing, no guessing.
How is this different from Google Shopping or a product feed?
Product feeds are designed for centralised indexes (Google, Bing, Facebook) that re-distribute your catalog. UCP discovery is designed for autonomous agents — the agent queries your store directly in real time, gets live stock and pricing, and can immediately create a cart (create_cart) against the same endpoint. No index, no lag, no handoff. It's discovery + transaction in one session.
Which AI agents can do UCP product discovery?
Any agent that speaks MCP. In practice: ChatGPT, Claude, Gemini, Apple Intelligence, and most frontier models as of 2026. UCP is transport-agnostic, but MCP is declared by ~100% of verified stores in our directory, so an MCP-capable agent works against essentially the entire ecosystem.
What capabilities does a store need to declare?
At a minimum: catalog-search or catalog-lookup. For the full discover → buy flow, agents also need cart, checkout, and at least one declared payment handler. Browse the full capability registry for the canonical list.
Where do I actually run discovery?
Three options depending on what you're trying to do. UCP Checker tells you whether a store supports discovery (validation layer). UCP Playground runs a real AI agent session against the store so you can watch every tool call in real time (no code required). UCP Registry exposes a live discovery API you can call from your own app (production scale). All three are designed to be used together: validate with Checker, test with Playground, integrate with Registry.
How often is this data updated?
Every store in our directory is re-crawled every 24 hours. The list of discovery-ready stores above updates automatically — any store that drops catalog-search or fails validation falls off on the next crawl cycle, any new store that adds it gets added.
Weekly UCP Report

Get the agentic commerce digest every Monday

Real adoption data, ecosystem trends, new spec versions, and the stores that broke or recovered this week. Read by founders and engineers building the next generation of commerce.

Free forever No spam Unsubscribe anytime

View a sample report →

Weekly UCP Report
Issue #54 · Oct 12, 2026
+664
new verified stores
Verified rate
Latest spec
Cart capability