Agent-ready checkout: the reference architecture
Last updated: · Leutrim Miftaraj
An agent-ready checkout consists of three layers: recognition (a legitimate agent? — TAP/attestation), authority (was it allowed to? — mandate verify) and execution (payment — AP2/SPT/agentic token). Building only layers one and three means accepting revenue with open liability.
The flow
Agent request with mandate header → drop-in middleware extracts the presentation → /v1/verify with action and amount → allow: proceed to payment (store the receipt); deny: machine-readable reason back to the agent; escalate: waiting state, the customer confirms via wallet push.
Idempotency and consumption
An idempotency key prevents double processing; after successful execution the consumption books the amount finally and returns the remaining limit — the basis for follow-up purchases under the same mandate.
Dispute readiness
Store the receipt ID with the order: in a dispute the signed, timestamped verification receipt is the answer — two-sided, not just from your own logs.
Key takeaway
Three layers, one principle: authority first, then the money.
Try it yourself
Issue your first agent mandate and check your first decision with its receipt — free, in a few minutes.