The Agent Payments Protocol (AP2) is a protocol for authorizing payments that AI agents make on a user’s behalf. Developer’s Guide to AI Agent Protocols describes it as adding typed mandates that provide non-repudiable proof of intent and enforce configurable guardrails on every transaction, so that an agent’s purchases carry a record of what limits were set, which merchants were approved, when the authorization expires, and who approved each payment.
Usage
The guide introduces AP2 at the point where an agent has already gained the ability to place orders and
the open question is who authorized the spending. Its flow has three typed objects. An
IntentMandate, configured by the owner, specifies the allowed merchants, conditions such as
refundability, whether cart confirmation is required, a spending limit for auto-approval and an expiry.
The agent then generates a PaymentMandate bound to a specific cart and amount; if the order exceeds
the limit, the mandate remains unsigned until a manager explicitly approves it. A PaymentReceipt
closes the audit trail. The guide summarizes the chain as recording what was intended, authorized and
paid.
AP2 is designed to work with the Universal Commerce Protocol (UCP): in the guide’s words, UCP handles what is ordered and from whom, while AP2 handles who approved the purchase and provides the audit trail, plugging into UCP as an extension that adds cryptographic proof of authorization to the checkout flow. The guide’s example simulates the manager’s signature and notes that real AP2 uses JWT or biometric signing on a secure device.
At the time of that post AP2 was at v0.1, and its types were provided as a separate package rather than built into the core of Agent Development Kit (ADK).
Related Terms
- Universal Commerce Protocol (UCP) — the commerce protocol AP2 extends
- Guardrails — AP2’s mandates are a protocol-level form of spending guardrail
- Human-in-the-loop — orders over the configured limit wait for explicit human approval