Directories check a paid tool's quote, never its checkout — this probe pays your endpoint the way a stranger would, with the official clients, and signs a verdict that the challenge is legible, the price matches in every dialect, and a real purchase settles and delivers.
A paid MCP/x402 endpoint can pass every listing check and still be unpayable: one seller was graded A, 14/14, on the morning it found a buyer using the official x402 v2 client could not pay it at all. The failure classes are checkout classes, not quote classes — the payment challenge returned as a JSON-RPC error that MCP clients swallow before the model ever sees it; the price written as decimal dollars in discovery and atomic units in the challenge; v1 and v2 header names and payload shapes diverging; the settlement receipt omitted so the buyer's books record $0 spent; quote headers too large for a proxy buffer. This service runs the test that catches them — the official-client stranger purchase, productised: (1) dialect probes across x402 v1 (body) and v2 (PAYMENT-REQUIRED header) and the MCP transports (JSON-RPC error, isError result, _meta); (2) a per-endpoint signed verdict — challenge LEGIBLE / SWALLOWED / DIALECT-PARTIAL, price CONSISTENT across discovery (decimal), challenge (atomic) and settlement, receipt PRESENT, delivery-after-settle CONFIRMED — where any cross-dialect price divergence settles nothing and grades the divergence; (3) continuous seller monitoring — daily unpaid dialect probes, weekly settled probes, a settled re-probe on every discovered price/contract change, and a regression alert when a deploy breaks a dialect; (4) a buyer-side verdict lookup a buying agent can query pre-commit, answered with settled evidence instead of a quote-side guess. The probe grades all 19 seeded fixture doubles correctly — all 17 catalogued HTTP failure modes and all 3 MCP legibility classes — at a measured mean probe time of 13.26 ms.
Stated plainly: the evidence is fixture evidence (19/19) — the live-endpoint catch rate is unmeasured; a passing verdict proves one official-client-shaped purchase settled and delivered at probe time, not that every client version will pay; willingness to pay for settled monitoring is unmeasured; on the settled leg the endpoint's own call price is a disclosed pass-through, stated before the run and bounded by a cap you set.
| Buyer verdict lookup | $0.01 per query — latest signed settled verdict per endpoint |
|---|---|
| Fresh settled probe, on demand | $0.05 per endpoint + the endpoint's own call price at cost (pass-through, disclosed pre-run, buyer-set cap) |
| Seller monitoring — single endpoint | $0.99/endpoint/mo — daily unpaid probes, weekly settled probe, change-triggered settled re-probes, regression alerts |
| Operator plan | $9.99/operator/mo — up to 20 endpoints, same cadence |
| Fleet plan | $19.99/operator/mo — up to 79 endpoints, daily unpaid probes, biweekly settled probes, change-triggered re-probes, regression alerts |
Settled purchases for endpoints priced above $0.05 are metered inside plans as pass-through against the registered cap. Trial: one free fresh settled probe per operator, ever. After that single probe, the standard prices apply to all lookups, probes, and plans.
API base: https://textforagents.com/products/paid-call-checkout-conformance-probe/ — start with POST /register, then POST /probe and GET /verdict. Docs, OpenAPI, and llms.txt resolve without auth at the API base. MCP tools (pricing) are served at POST /mcp on the API base; paid work collects on the HTTP x402 surface. Support: support@textforagents.com
Provided by TextForAgents LLC.