A store can carry a 97/100 “UCP Score” while its checkout capability has never been probed by anything live. That’s not a bug in UCP Checker — it’s what the number is built to measure. Its own methodology page, fetched for this piece, says so directly: “by default the score is computed from static signals only.” Understanding that split matters right now, because the same September 2026 census that reports the number also shows a live example of the gap it doesn’t close.

What the 0-100 number actually weighs

UCP Checker’s score is a composite of three sub-scores: Agent Discovery (30%) checks HTTPS, reachability, robots.txt, and surface signals like /llms.txt and Organization JSON-LD. UCP Conformance (40%, the largest share) checks whether the manifest at /.well-known/ucp parses and validates against the spec — an invalid manifest caps this category near 50. Capability Coverage (30%) rewards declared transports, checkout support and payment handlers.

All three run from a single unauthenticated fetch of the manifest file. The methodology is explicit that this is deliberate: “fast, cheap, safe for daily-cron-scale checking” across the more than 22,000 storefronts it re-crawls. A separate, opt-in layer — triggered only when someone manually runs a check — adds read-only “functional probes” that ping each declared transport endpoint (an MCP tools/list, a REST GET) to see if it answers at all. Even then, the page is explicit about the boundary: “Probes are deliberately read-only… they do not exercise checkout or write any state.” Actually completing a purchase is handled by a separate product, UCP Playground, which the same page describes as answering “whether agents can actually complete a transaction” — a different question from the score.

The gap, live, in the same report

The census that ships alongside this methodology catches the gap in the act. On September 22, Wix storefronts began declaring the dev.ucp.shopping.cart capability fleet-wide; by mid-afternoon UTC, 3,549 of 6,929 Wix stores (about 51%) had it. Cart and order are genuinely separate capabilities in the UCP specification — Cart is explicitly scoped to “basket building… no payment configuration needed,” with “no status lifecycle” and no complete operation, while Order represents “confirmed transactions resulting from a successful checkout submission.” Wix’s rollout added only the first. Per the census: “Order is still at zero across the Wix fleet, and no Wix store has moved to 2026-08-25” — the spec version Shopify, whose stores already declare cart and order at 100%, adopted a month earlier.

None of that shows up as a Capability Coverage penalty for an ordinary, un-probed Wix listing, because a store that doesn’t declare order isn’t “failing” a checkout test — it’s correctly reporting that it hasn’t built one yet. The manifest is honest; it just isn’t the same claim as “an agent can buy here.”

Declared isn’t delivered elsewhere too

The same census names other platforms where the declaration and the capability diverge in ways a manifest-only score can’t catch: BigCommerce stores declare checkout, cart and order but no catalog capability; Zid and Tray publish catalog-only profiles (discoverable, not purchasable); and five storefronts on the Czech platform Shoptet are “publishing, not yet valid” because their profiles omit the required payment_handlers field. Each is a legitimate, different failure mode a merchant’s own manifest can hide — which is the same reason a 96-vs-41 score split between two scanners isn’t necessarily either one being wrong: they’re often reading different layers of the same store.

FAQ

Does UCP Checker’s score mean my checkout works for AI agents?

No, not by default. The score’s Capability Coverage component rewards declaring checkout, payment handlers and transports; only when a user manually triggers a check do read-only probes confirm a declared endpoint responds at all, and even those probes explicitly don’t attempt a purchase. Whether a transaction actually completes is answered by UCP Checker’s separate Playground product, not the score.

What’s the difference between UCP’s Cart and Order capabilities?

Cart (dev.ucp.shopping.cart) is pre-purchase basket building with no payment step and no completion state. Order (dev.ucp.shopping.order) represents a confirmed transaction after checkout — line items, fulfillment and post-purchase adjustments. A store can declare one without the other, as Wix’s September 2026 rollout did by adding cart fleet-wide while order stayed undeclared.

Why would a store score high on UCP Conformance but still not be agent-purchasable?

Conformance (40% of the score) only tests whether the manifest is well-formed JSON that validates against the spec’s required fields — it says nothing about which capabilities are declared. A store can publish a perfectly valid manifest that declares catalog search only, scoring well on Conformance while offering an agent no way to complete a purchase.

Sources

Scores and manifests both describe a store’s intentions; the only way to know if an agent can actually finish a purchase is to test the transaction itself, which is the layer AgentReady’s own scan is built around.