Agent audit records
The MCP audit event includes the agent, entity, vault, grant, tool name, category, correlation identifier and policy version. Outcomes includeallow, deny, step_up_required and error.
The event hashes the input and records selected fields, such as amount and chain. It does not store a full copy of every input and output. Audit delivery depends on the configured audit sink; a successful tool response alone does not prove that an audit event was stored.
Payment receipts
Where a payment flow produces a receipt, review its payment identifier, status, amount and network reference. Use an on-chain transaction hash to inspect the corresponding transaction on a block explorer. An audit event is not a substitute for that settlement check. See the Receipt schema and Agent activity event schema for their respective formats.View and export activity
Use the activity views available for your account. The current agent audit page does not provide a connected live stream, replay filters or working JSON/PDF export buttons. Do not interpret its empty state as proof that no actions occurred. Owner and Admin access can query entity-scoped agent records through the forensics export API. It returns JSON data with optional agent and date filters. A signed PDF and a fixed export completion time are not provided by that API. For an MCP integration,audit.stream defines the cursor contract. A usable stream also requires the deployment to provide the corresponding SSE endpoint.
Tamper-evidence and failures
Append-only records and hashes help with later review. They do not reverse an external transfer. If receipt or audit delivery fails, verify the payment’s status before retrying.Retention
The retention model defines hot, warm, cold and optional regulatory tiers. In the v0.35 source, the scheduled sweep returnsretention_enforcement_not_deployed; it does not move or delete rows. An age threshold does not prove physical archival or deletion. See Retention tiers.