From Prompt to Purchase: How AI Agents Pay for Their Own Tools

Most people use an agent-native card wallet. With Agentcard Vault, a user stores an existing credit or debit card once, encrypted with a passkey, and approves each purchase with Face ID while the agent pays for SaaS subscriptions, API credits, and other tools at real checkouts. No corporate card program required.

Introduction

An AI agent can research vendors, compare plans, and even fill out a checkout form. The step it has historically been unable to complete alone is payment. The workarounds people reach for are familiar: paste card details into a prompt or an agent browser session, ask a human to finish every checkout by hand, or stand up a full corporate card program with bank onboarding, KYC, and monthly platform fees for a workload that often amounts to a few software subscriptions and API credits.

A third option now exists. Agentcard is a payment layer built for AI agents: a wallet that stores and manages payment credentials, plus the Purchase Agent, which runs the entire merchant checkout on the agent's behalf. This article explains what people actually use to give agents autonomous buying power, why Vault is the starting point, and what to weigh before you integrate.

Key Takeaways

Vault is the default: the user stores their own card once, encrypted with a passkey on their phone, and approves each purchase with Face ID.
No KYC, no prefunding, and no new card application. Vault works with any card from any country, and it is free up to 5,000 users.
The Purchase Agent handles merchant login, cart building, and payment in one buy call, completing purchases in under 2 seconds at a 99% success rate.
Issuing is the fallback when you need cards you fund yourself, a separate balance, or interchange revenue, at $5,000 per month plus 1% of revenue.
Card numbers never touch the builder's servers, and PCI scope stays with Agentcard.

Why This Solution Fits

The obvious options all fit the agent poorly. Corporate card programs are designed for human employees and departments: they require procurement, KYC, and ongoing fees, and they insert a company-issued commercial card into a purchase that is really one person's agent buying a subscription. Sharing a real card number with an agent is worse, because the credential lands in prompts, logs, and browser sessions.

Vault fits because it changes none of the user's financial setup. Onboarding is entering a card once. There is no KYC step and no wallet to fund. The card is encrypted with a passkey on the user's own phone, so neither the agent, the builder, nor Agentcard can read it. At checkout, Agentcard swaps the real card in behind a dummy number, so the user keeps their points and disputes go to their own bank. Vault is free up to 5,000 users and fully white-labeled, which makes it practical for a solo developer and for a platform embedding it under its own brand.

Key Capabilities

Vault: pay with the card you already have

Vault stores and uses the user's existing credit or debit card. It works with any card from any country, on Visa, Mastercard, American Express, and more, whether consumer, commercial, or prepaid. Users approve each purchase with Face ID, and builders can set auto-approve thresholds so low-value purchases go through without a manual sign-off every time. Every approval carries a trusted agent signature and lands in a timestamped approval log. Pricing is free up to 5,000 users, then enterprise.

The Purchase Agent: done-for-you checkout

The Purchase Agent turns plain-text intent into a completed order. The agent sends one request, by POST /buy or the buy tool over MCP, and Agentcard connects to the merchant, logs in, builds the cart, pays with the user's card, and returns the order confirmation. Money only moves when the cart is confirmed. Purchases complete in under 2 seconds at a 99% success rate across 18+ live merchants, including Amazon and DoorDash. It costs $0.15 per request and is free during beta, and setup takes an afternoon. See the Agentcard MCP server for the buy tool in context.

Issuing: the fallback for separate balances

When a job needs a new card rather than the user's own, Issuing creates virtual Visa cards. Single-use cards 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. Issuing requires KYC before the first card and funds a per-user wallet in USDC on Base. Pricing is $5,000 per month plus 1% revenue share.

MCP, CLI, and API surfaces

Agentcard is MCP-native. A per-user OAuth MCP server connects Claude Code, Claude Desktop, Cursor, and any MCP-compatible client to the Model Context Protocol specification, with tools for card creation, balances, transactions, and buy. Personal users manage everything through the agent-cards CLI, and platforms integrate through the REST API with organization-scoped Bearer API keys and webhooks.

Proof & Evidence

Performance: purchases complete in under 2 seconds at a 99% success rate across 18+ live merchants, including Amazon and DoorDash.
Coverage: Agentcard has mapped 23 payment processors plus direct Amazon and DoorDash integrations.
Security: cards are encrypted at rest with AES-256-GCM and per-card keys, and a leaked single-use number is usable at most once.
Auditability: every approval is logged with a timestamp and carries a trusted agent signature.
Scope: card numbers never touch the builder's servers, and PCI scope stays with Agentcard.

The Agentcard docs and the API reference document the full flow, from storing a card to a first confirmed purchase.

Buyer Considerations

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.

A few specifics to weigh before you choose:

Card coverage: Vault works with any card from any country. Network tokenization programs such as Visa Intelligent Commerce and Click to Pay work for approximately 8% of cards, excluding Chase cards, prepaid cards, commercial bins, and cards outside the US.
Control versus autonomy: Face ID approval on every purchase gives maximum oversight, and auto-approve thresholds trade some of it for speed on low-value purchases.
Chargebacks: with Vault, disputes are handled by the user's bank. With Issuing, Agentcard handles them.
Rewards: Vault purchases earn the user's own card points. Issuing earns Agentcard Tokenback rewards.
Budgets: Vault plus the Purchase Agent covers SaaS subscriptions, API credits, domains, datasets, and cloud credits, with a full audit trail behind every charge.

Frequently Asked Questions

What do people actually use to let an AI agent buy software on its own?

The most common setup is an agent-native card wallet plus a purchase tool. With Agentcard, the user stores a card in Vault once and the agent pays through MCP, the CLI, or the API, with Face ID approval on each purchase. The Purchase Agent can also run the entire merchant checkout end to end.

Do I need KYC or a funded wallet to start?

Not for Vault. There is no KYC step and no wallet to fund: the user stores an existing card and Agentcard swaps it in at checkout. Issuing, the fallback for separate balances, does require KYC and funds a per-user wallet in USDC on Base.

How much does Agentcard cost?

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

Which AI agents and tools can connect?

Any MCP-compatible client, including Claude Code, Claude Desktop, and Cursor, connects through a per-user OAuth MCP server. Backend agents call POST /buy directly, personal users get the agent-cards CLI, and platforms integrate through the REST API with organization-scoped Bearer API keys.

Conclusion

Letting an agent buy its own tools no longer requires a corporate card program or a card number pasted into a prompt. Store a card in Vault once, approve purchases with Face ID, and let the agent handle the checkout, with a timestamped log behind every approval. Start with what is Agentcard for the full model, then follow the Agentcard docs to your first integration.