The diagnostic runs 38 checks across the six criteria. Each criterion gets a score from 0 to 100, and the overall score is a weighted average: catalog 25%, cart 20%, checkout 15%, bot policies 15%, payment 15% and order traceability 10%. From 85 points the store is ready for agents; from 70, ready with some gaps; from 40, partly ready; below that, not ready yet. If the store blocks the reading, the report says so instead of showing a low score. The report shows what was found and the five fixes with the greatest impact.
01
Agent-accessible catalog
An agent needs to understand what the store sells without interpreting the page design. Product, price and stock data must be machine-readable, such as schema.org Product and Offer, catalog feeds or an up-to-date sitemap. If the offer exists only as an image or loose text, the agent has no reliable basis to compare and choose.
13 checks: robots.txt, XML sitemap, llms.txt, Product and Offer JSON-LD, variants, catalog feed, price in the initial HTML, Organization and sameAs, trust pages, metadata, indexability and home page size.
02
Cart endpoints
Choosing the product is only the start: the agent needs to build a cart. That can be a cart API or predictable URLs that always lead to the same set of items. Without a stable path, the visit stops at the storefront and the order never exists.
4 checks: identifiable platform, public product JSON, cart read (such as /cart.js) and price with variant or availability.
03
Agent-compatible checkout
Checkout needs to be a flow an agent can complete, not just a sequence of clicks designed for a person. That includes compatibility with agentic commerce protocols and a clear path to order confirmation. If a step only works with human interaction on screen, the purchase stops halfway.
5 checks: captcha or challenge, guest checkout, UCP profile at /.well-known/ucp, ACP document at /.well-known/acp.json and other machine-to-machine channels.
04
Bot policies
The store needs to declare which automations it accepts. This shows up in robots.txt, llms.txt and firewall or bot management rules: they shouldn’t block legitimate agents and should be able to identify verified bots. Treating all automated access as abuse also shuts the door on those who would come to buy.
9 checks: robots.txt rules per agent, home page response for ChatGPT-User, OAI-SearchBot, Claude-User and PerplexityBot, probe access, HTTP redirects, content without login, Markdown response, useful 404, training-crawler policy and signed bots.
05
Order traceability
After payment, order status and tracking need to be queryable by machines. The agent, or the person who authorized it, must be able to follow the order without copying codes from an email into another site. A status lookup via API or structured data closes the sales loop.
3 checks: tracking page without an account, Order or ParcelDelivery JSON-LD and a named carrier.
06
Payment capability
The purchase only ends when there is a payment method the agent can use and authorize. That includes programmatic payments, where the charge follows a rule agreed in advance, without someone typing in details for every order. If the only path is a button made for a person, this step remains incomplete.
4 checks: payment methods written in the HTML (Pix, card, boleto), Pix, identifiable gateway and programmatic payment: UCP with payment handlers, ACP or /.well-known/x402.