REFERENCE · STABLE DEFINITIONS
Agentic Payments Glossary
Precise definitions for AI-agent payments, delegated purchase authority, payment protocols, execution reservations and consume-once controls.
Reviewed: July 24, 2026 · “Authorization” and “authorisation” refer to the same policy concept on this site.
Why these terms need precise boundaries
Identity, intent, authorization, payment execution and settlement are different layers. Collapsing them into “the agent paid” hides the exact control that failed. These definitions describe the product's implemented semantics and link protocol names to their practical payment boundary.
Agentic commerce
Commerce in which software agents can discover, compare, configure or purchase goods and services on a person's or organization's behalf.
AI-agent payment
A payment attempt selected or initiated by an AI agent. The agent's ability to call a payment tool does not itself establish delegated authority.
Mandate
In MandateShield, an immutable version of registered AI-agent purchase policy. It is not a SEPA, ACH or Bacs direct-debit instruction.
Merchant binding
An exact comparison that prevents the final seller or payee from being substituted after authority was granted.
Cumulative budget
A lifetime, daily or monthly cap shared across qualifying purchase attempts. Correct enforcement reserves spend atomically so concurrent agents cannot each observe the same remaining balance.
Execution reservation
Short-lived, persisted authority created by qualifying strict verification. It reserves budget but is not a payment instruction, provider submission or settlement record.
PROCESSOR consumption
The audience-bound transition by which a trusted payment gateway atomically claims one reserved authorization before exactly one provider submission.
Decision receipt
A signed, independently verifiable record of the verification decision, bindings and evidence hashes. It proves what MandateShield decided, not that a processor settled money.
Replay tombstone
Durable account-wide state recording that an execution attempt identifier was already used, preventing later duplicate authorization even after a short reservation expires.
MPP — Machine Payments Protocol
A payment-method-agnostic HTTP 402 authorization framework for machine payments. Payment protocol validity does not by itself prove delegated purchase authority.
x402
An HTTP-native payment protocol in which a service returns payment requirements and a client supplies an onchain payment payload before the resource is delivered.
AP2 — Agent Payments Protocol
An open protocol for interoperable agent-led payments that uses cryptographic mandates to carry user intent and transaction evidence.
UCP — Universal Commerce Protocol
An open protocol for interoperable commerce capabilities, checkout and services. Interoperability does not eliminate execution-time authorization checks.
ACP — Agentic Commerce Protocol
A protocol for agent-compatible product discovery and checkout while the merchant remains responsible for payment processing and fulfillment.
TAP — Trusted Agent Protocol
Visa's protocol for recognized-agent signals and signed intent. Agent identity is an input to authorization, not proof that every amount and seller was approved.
MCP — Model Context Protocol
An open protocol through which AI applications discover and call tools or access contextual resources. Tool access must remain separate from payment authority.
A2A — Agent2Agent Protocol
A protocol for agents to discover capabilities and coordinate tasks across systems. A2A transport does not weaken purchase-policy or consume-once requirements.
Continue with implementation evidence
Use the payment-authority control matrix for claim-by-claim product evidence, the protocol support matrix for exact adapter boundaries and the MCP payment authorization guide for a safe agent integration sequence.