Methodology

How the score works

The check plays the part of a shopping agent that has no browser and no account. It requests public pages and well-known files, reads what comes back, and scores what it can observe. It never signs in, adds to a cart or places an order.

What a scan does

One scan sends about 30 requests, four at a time, with a clearly labelled user agent. It reads the homepage, robots.txt, the sitemap and llms.txt, finds one representative product page, and probes the well-known locations where commerce and agent protocols publish their discovery documents. Where a store declares a Model Context Protocol endpoint, the scan performs the handshake and lists the tools. It does not call them.

Scoring

Checks are grouped into four stages that follow an agent's journey. Each check carries a weight. A pass earns the full weight, a partial result earns part of it, and a miss earns nothing. A stage score is the share of weight earned in that stage. The overall score is the weighted average of the four stage scores.

Checks marked emerging cover standards that are still gaining adoption. They carry less weight and are never listed ahead of the fundamentals. Checks that cannot be assessed are left out of the score. Product checks count as missed when no product page can be reached, because an agent in the same position could not read the product either.

Limits

The score is an automated snapshot, not a certification, a security review or a guarantee that an agent can or cannot complete a purchase. Results reflect one product page and one moment in time. Stores that vary content by market or protect pages against automated traffic can score differently from one scan to the next.

Some integrations cannot be seen from outside. The Agentic Commerce Protocol relies on a product feed and checkout endpoints shared privately with the assistant, and card-network agent programmes verify signatures at the merchant's edge. A store can support these and still show nothing to this check.

Sources

Checks follow published specifications: RFC 9309 for robots.txt, schema.org and Google's merchant listing guidance for product data, the Universal Commerce Protocol, the Model Context Protocol, the Agent2Agent protocol, RFC 9727 for API catalogues and RFC 9728 for OAuth resource metadata. Each check below links to its reference. The full catalogue is available as JSON.

The idea of scoring agent readiness with evidence and fixes follows Is Agentic by Vercel and Ora, which scores general agent readiness. This check uses its own scanner and its own method, focused on commerce. Where Is Agentic has a stored report for the same site, we link to it for comparison.