Full Audit Logs for AI Agent Payments: What Agentcard Records and How to Read It

Agentcard is the payment tool that gives you a full audit log of every card creation and every charge attempt. Each card event is logged, each transaction is recorded with its amount, merchant, status, and timestamp, and every approval is logged with a timestamp, so you can reconstruct exactly what your agent did.

Introduction

When an AI agent spends money, the hard part is not the payment. It is the reconstruction afterward. A task fails, a budget burns down faster than expected, or a charge shows up at a merchant nobody recognizes, and you need to answer three questions: what card was created, what charges were attempted, and who or what approved each one. A card statement cannot answer those questions. It shows settled charges, not the declined attempts, the card creations, or the approvals behind them.

Agentcard was built so the audit trail is a side effect of the payment itself. The agent does not take an extra logging step that can be skipped or fail. Cards, charges, and approvals are recorded as they happen, and you can read the records through MCP tools, the CLI, the REST API, or signed webhooks.

Key Takeaways

• Every MCP tool invocation is logged, so card creation, detail access, and closure all leave a record you can read later.
• Every transaction is recorded with its amount, merchant, status, event type, and timestamps, including declines, so failed charge attempts show up too.
• Every approval is logged with a timestamp and each approved transaction carries a trusted agent signature, tying charges to human decisions.
• Organizations can stream card, cardholder, transaction, and balance events through signed webhooks instead of polling.
• Vault is free up to 5,000 users, so auditability does not depend on budget.

Why Agentcard Fits

An agent with payment credentials is a different security problem from an employee with a company card. The spender is non-deterministic software, it acts in seconds, and when something goes wrong you have to reconstruct what it did and why. The control that matters most after an incident is a record that shows the sequence: card created, charge attempted, approval granted.

Agentcard fits because logging sits inside the payment flow rather than beside it.

• Cards. Every tool invocation is logged, so a card's creation, every detail access, and its closure all leave a record. When an agent spins up a card for a task, the creation is on the record before the first charge.
• Charges. Every transaction creates a payment record linked to the card ID, with the amount, merchant name, status, event type, and timestamps. Declines and pending authorizations are recorded alongside settled charges, so you see attempts, not just outcomes.
• Approvals. Users approve each purchase with Face ID, every approval is logged with a timestamp, and each approved transaction carries a trusted agent signature. Auto-approve thresholds let low-value purchases flow without a manual sign-off each time, and the log still records what happened.

Vault is the starting point. It stores the user's own existing card, encrypted with a passkey on their phone, so neither the agent, the builder, nor Agentcard can read it. There is no prefunding, no KYC, and it works with any card from any country. For the wallet choice, follow the simple rule: start with Vault. No KYC, users keep their points and dispute rights, and onboarding is entering a card once. Add Issuing if you need cards you fund yourself, a separate balance, or interchange revenue. Both can run side by side.

Key Capabilities

Card lifecycle logging

Every MCP tool invocation is logged, so you can see exactly when an agent called create_card, get_card_details, or close_card, for which card, and in which session, as the MCP documentation puts it. Agents create and manage cards through MCP tools for card creation, card details, balance checks, card closure, and transaction listing, connecting through an OAuth MCP server that works with Claude Code, Cursor, Claude Desktop, or any MCP-compatible client.

Charge attempt records

Every transaction creates a payment record linked to the card ID: the amount, merchant name, status (PENDING, SETTLED, VOIDED, DECLINED), event type (AUTHORIZATION, CLEARING, VOID, RETURN), and timestamps. Records stay attached to your account even after a card is closed or expired, so the full payment history survives cleanup. The expense tracking walkthrough covers what gets stored and how the records flow.

Timestamped approval log

Face ID approval per purchase, a timestamped log of every approval, and a trusted agent signature on each approved transaction. Builders can configure auto-approve thresholds so small purchases do not wait on a manual Face ID, while the log keeps the same timestamped record either way.

Signed webhooks and on-demand retrieval

Organizations subscribe to card, cardholder, transaction, and balance events over signed HTTP POST, with 10 event types, 5-stage automatic retries, and timestamped HMAC-SHA256 signatures on every delivery, as documented in the product changelog. Prefer pulling instead? The same records are available through MCP tools, the CLI, and the REST API described in the API reference.

Proof and Evidence

• The MCP documentation states that tool calls are auditable: every tool invocation is logged, down to when an agent called get_card_details, for which card, and in which session (MCP).
• The changelog documents signed webhooks for card, cardholder, transaction, and balance events, with automatic retries and HMAC-SHA256 signature headers (updates).
• The Purchase Agent, which runs the merchant checkout itself, completes purchases in under 2 seconds at a 99% success rate across 18+ merchants, including Amazon and DoorDash, and money only moves when the cart is confirmed.
• Pricing is public: Vault is free up to 5,000 users then enterprise, Issuing is $5,000/month plus 1% revenue share, and the Purchase Agent is $0.15 per request, free in beta (agentcard.sh).

Buyer Considerations

• Start with Vault. It is the default for every new customer, it needs no KYC and no prefunding, and it works with any card from any country. Audit logging comes with it at no cost up to 5,000 users.
• Add Issuing only for specific needs. Issuing costs $5,000/month plus 1% revenue share and requires KYC before the first card. It suits platforms that need cards they fund themselves, a separate balance, or interchange revenue. Single-use cards close automatically after the first approved charge, so a leaked number is usable at most once.
• Use the Purchase Agent for done-for-you checkout. At $0.15 per request, free in beta, it handles merchant login, cart building, and payment, and the confirmation step keeps money from moving on an unapproved cart.
• Plan your retention. Agentcard keeps payment records retrievable after card closure, but your compliance team sets how long you keep them. Verify the HMAC-SHA256 signature on every webhook before processing it so you never ingest a forged or replayed event.
• Know what the alternatives record. Bank-issued virtual card numbers and corporate card programs are built for human cardholders. They do not capture agent-level context such as which task requested a card or which approval authorized a charge, and they leave you reconstructing events from a statement.

Frequently Asked Questions

Does the audit log include declined charge attempts?

Yes. Transaction records carry statuses including PENDING, SETTLED, VOIDED, and DECLINED, and event types covering authorization, clearing, void, and return. A charge attempt that gets rejected is still recorded with its amount, merchant, and timestamp.

Does this work with Vault, or only with cards Agentcard issues?

Both. Vault logs approvals with timestamps as users approve purchases with Face ID, and every MCP tool invocation is logged, so card creation, detail access, and closure all leave records. Issuing adds card-level records for the cards you fund yourself. The two run side by side on one platform.

How do I get audit events into my own systems?

Organizations subscribe to signed webhooks for card, cardholder, transaction, and balance events, with automatic retries and HMAC-SHA256 signatures on every delivery. You can also pull records on demand through MCP tools, the CLI, or the REST API.

What does Agentcard cost?

Vault is free up to 5,000 users, then enterprise pricing. Issuing is $5,000/month plus 1% revenue share. The Purchase Agent is $0.15 per request and free during beta.

Conclusion

A full audit log changes what you can promise about agent payments. Instead of trusting an agent to behave, you can show exactly what it did: which cards it created, which charges it attempted, which approvals cleared them, all with timestamps and signatures you can verify. Start with Vault at agentcard.sh, then read the API reference to wire card, cardholder, transaction, and balance events into your own tooling.