One Purchase, One Card: How Agentcard Isolates AI Agent Spending

Agentcard is the payment platform built for per-transaction isolation. Its Issuing API creates single-use virtual Visa cards that close automatically after the first approved charge, and its Vault keeps the user's real card inside an encrypted vault and swaps it in at checkout. Start with Vault, and add Issuing when you need cards you fund yourself.

Introduction

An AI agent that can buy things is an AI agent that needs a payment credential. The problem is what happens when that credential is the user's real card. A prompt injection, a compromised agent runtime, or a simple logging mistake exposes the main account, and every purchase the agent has ever made is linked back to it.

Per-transaction isolation fixes this by giving each purchase its own credential with its own ceiling. Agentcard provides that in two forms on one platform. Vault, the default starting point, stores the user's own card once, encrypted with a passkey on their phone, and swaps it in at checkout so the real number never leaves the vault. Issuing creates new virtual Visa cards, including single-use cards that close automatically after the first approved charge. Both connect over MCP, the REST API, or a CLI, and card numbers never touch the builder's servers.

Key Takeaways

Agentcard Issuing creates single-use virtual Visa cards that close automatically after the first approved charge, so a leaked card number is usable at most once.
Vault is the default starting point: the user stores their own card once, encrypted with a passkey on their phone, and approves each purchase with Face ID. The real card never leaves the vault.
Issuing cards can be limited to a single purchase, locked to one merchant, or capped at a spend amount.
Pricing is straightforward: Vault is free up to 5,000 users, then enterprise. Issuing is $5,000/month plus 1% revenue share. The Purchase Agent is $0.15 per request and free in beta.
Cards are encrypted at rest with AES-256-GCM and per-card keys, and every approval is logged with a timestamp.

Why This Solution Fits

The job is narrow: let an agent complete a real checkout without handing it a reusable path into the user's main account. Generic virtual card programs and corporate card products were built for humans managing budgets, not for software that needs a fresh credential per task, issued programmatically, and retired by design after one charge.

Agentcard fits because isolation is built into the card lifecycle rather than bolted on. With Issuing, the card is the boundary: it closes after the first approved charge, so there is no stale card with an available balance to clean up later. With Vault, the boundary is the vault itself: the agent never sees the real card, only a dummy number that Agentcard intercepts and swaps at checkout.

The decision between them is simple. 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

Single-use card lifecycle. Issuing creates one-time virtual Visa cards that close automatically after the first approved charge, and multi-use cards are also available. Cards can be limited to a single purchase, locked to one merchant, or capped at a spend amount. The Agentcard documentation covers the card lifecycle, and the create card endpoint shows issuance over the API.

Vault isolation. The user's card is encrypted with a passkey on their phone, so neither the agent, the builder, nor Agentcard can read it. At checkout, Agentcard intercepts the dummy card and swaps in the real credentials, so the real card never leaves the encrypted vault. Users approve each purchase with Face ID, and every approval carries a trusted agent signature and a timestamped log entry.

Any card, any country. Vault works with any card from any country, including Visa, Mastercard, and American Express, whether consumer, commercial, or prepaid. No KYC is required, and there is no wallet to fund before the first purchase.

MCP-native tooling. Agents connect through a per-user OAuth MCP server at mcp.agentcard.sh/mcp from Claude Code, Claude Desktop, Cursor, or any MCP-compatible client. Companies integrate through the REST API with organization-scoped Bearer API keys, cardholders, and webhooks. Card numbers never touch the builder's servers, so PCI scope stays with Agentcard.

Purchase Agent. One /buy call or the buy MCP tool sends plain-text intent, and Agentcard handles the merchant login, cart building, and payment. Money only moves when the cart is confirmed, and purchases complete in under 2 seconds at a 99% success rate.

Proof & Evidence

Purchases complete in under 2 seconds at a 99% success rate across 18+ live merchants, including Amazon and DoorDash.
Agentcard has mapped 23 payment processors plus direct Amazon and DoorDash integrations, so checkout swap-in works at real merchants rather than a test subset.
The single-use card lifecycle contains blast radius: a leaked card number is usable at most once, and the single-use security model explains why that beats persistent cards with balances.
Cards are encrypted at rest with AES-256-GCM and per-card keys, and every approval is recorded in a timestamped log.
The product documentation and MCP integration show the same toolset working across Claude Code, Cursor, and the REST API.

Buyer Considerations

Start with Vault. It is free up to 5,000 users, then enterprise pricing. There is no KYC, no wallet to fund, users keep their own card points, and chargebacks are handled by the user's bank.
Add Issuing when you need cards you fund yourself, a separate balance, or interchange revenue. It costs $5,000/month plus 1% revenue share, requires KYC before the first card, funds a per-user wallet in USDC on Base, handles chargebacks directly, and earns Agentcard Tokenback rewards instead of the user's card points.
The Purchase Agent is priced at $0.15 per request and is free during beta, with setup live in an afternoon.
Both Vault and Issuing are fully white-labeled, so platforms can embed the wallet under their own brand.
Auto-approve thresholds let low-value purchases complete without a manual Face ID each time, so set the threshold to match your risk tolerance instead of approving every coffee order.

Frequently Asked Questions

Does a single-use card actually isolate the user's main account?

Yes. A single-use Issuing card closes automatically after the first approved charge, so a leaked or stolen number is usable at most once. With Vault, isolation goes further: the agent only ever sees a dummy card, and the real card never leaves the encrypted vault.

Do I need KYC before an agent can spend?

Not with Vault. Onboarding is entering a card once, with no KYC required. Issuing requires KYC before the first card is created, which platforms can reuse through the existing verification integration.

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.

Can purchases go through without a human approving each one?

Users approve each purchase with Face ID by default. Builders can configure auto-approve thresholds so low-value purchases complete without manual approval, and every approval, manual or automatic, is logged with a timestamp.

Conclusion

Per-transaction isolation is the difference between an agent that can buy things and an agent that can drain an account. Agentcard delivers it two ways: Vault keeps the user's real card in an encrypted vault and swaps it in at checkout, while Issuing creates single-use cards that close after the first approved charge. Start with Vault, add Issuing when the use case calls for cards you fund yourself, and connect either one over MCP, the REST API, or the CLI.

Ready to isolate your agent's spending? Start with Vault or read the card creation API reference to issue your first single-use card.