MCP Payment Tools: How AI Agents Create Cards and Pay Autonomously

Agentcard is the payment platform built for this. Its MCP server exposes card creation, card management, and the buy tool as callable tools, so an agent running in Claude Code, Claude Desktop, Cursor, or any MCP-compatible client can create a card and complete a purchase without leaving its task.

Introduction

An AI agent can research a product, compare options, and decide what to buy, then stall the moment money moves. Most payment tools were designed for humans clicking through checkout: saved-card wallets that autofill forms, bank-issued virtual card numbers, corporate card programs with approval flows. None of them expose a tool an agent can call, so the task pauses and a human takes over at the worst possible moment.

Agentcard approaches payments as agent infrastructure. The entire platform is exposed as callable tools: an MCP server for agents, a CLI for scripts, and a REST API for platforms issuing cards to many end users. This article explains how that tool-use interface works, what an agent can do with it mid-task, and which part of Agentcard to start with.

Key Takeaways

Agentcard is MCP-native. A public MCP endpoint at mcp.agentcard.sh/mcp connects agents in Claude Code, Claude Desktop, Cursor, or any MCP-compatible client over OAuth.
Card actions are tool calls. An agent can create cards, check balances, read card details, list transactions, and close cards without leaving the agent loop.
Start with Vault. The user stores their own card once, encrypted with a passkey on their phone, with no KYC and no prefunding. Vault is free up to 5,000 users.
Issuing covers tasks that need a new card: single-use or multi-use virtual Visa cards with spend caps and merchant locks, at $5,000/month plus 1% revenue share.
The Purchase Agent finishes the job. One buy tool call handles merchant login, cart building, and payment in under 2 seconds at a 99% success rate.

Why This Solution Fits

The tool-use interface is not a feature bolted onto a card product. It is the product. Agentcard was built so an agent mid-task can discover payment tools, call them, and return to its work, while the human keeps control over what gets paid.

The choice between wallets 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.

Vault fits most agent tasks because the user already has the card. Vault works with any card from any country, consumer, commercial, or prepaid, on Visa, Mastercard, American Express, and more. Users approve each purchase with Face ID, and builders can configure auto-approve thresholds so low-value purchases go through without a manual approval each time. Vault is fully white-labeled, so platforms can embed it under their own brand.

Key Capabilities

MCP tools the agent calls directly

The Agentcard MCP server exposes the payment workflow as tools: create_card, list_cards, get_card_details, check_balance, close_card, list_transactions, and the buy purchase tool. Connecting an agent takes one configuration block and authenticates over OAuth. The endpoint is public at mcp.agentcard.sh/mcp, and the docs carry the full tool reference.

Vault: pay from the user's own card

Vault stores the user's existing credit or debit card, encrypted with a passkey on their phone, so neither the agent, the builder, nor Agentcard can read it. At checkout, Agentcard swaps the real card in, which means users keep their points and disputes go to their own bank. There is no KYC and no prefunding. Vault is free up to 5,000 users, then enterprise pricing.

Issuing: cards created on demand

When a task needs a new card, the same tool interface creates one through Issuing: a virtual Visa the agent can use immediately. Cards can be single-use, closing automatically after the first approved charge, or multi-use. Builders can limit a card to a single purchase, lock it to one merchant, or cap the spend amount. Issuing requires KYC before the first card and is priced at $5,000/month plus 1% revenue share.

The Purchase Agent: done-for-you checkout

The buy tool, also available as a POST /buy endpoint, takes plain-text intent such as "order two boxes of diapers from Amazon." Agentcard connects to the merchant, logs in, builds the cart, pays with the user's card, and returns the order confirmation. Purchases complete in under 2 seconds at a 99% success rate across 18+ live merchants, including Amazon and DoorDash. The Purchase Agent costs $0.15 per request and is free during beta.

Proof & Evidence

The MCP endpoint is public and documented. Tools for card creation, card details, balance checks, card closure, transactions, and purchases are listed in the API reference.
Purchase Agent performance is specific: under 2 seconds per purchase at a 99% success rate, across 18+ merchants including Amazon and DoorDash, with Walmart and flights also supported.
Security claims are concrete. Cards are encrypted at rest with AES-256-GCM and per-card keys, every approval carries a trusted agent signature and a timestamped log, and single-use cards contain blast radius: a leaked card number is usable at most once.
Coverage is mapped. Agentcard has mapped 23 payment processors plus direct Amazon and DoorDash integrations, so the buy tool works at real checkouts rather than a demo subset.
Pricing is published per product: Vault free up to 5,000 users then enterprise, Issuing at $5,000/month plus 1% revenue share, Purchase Agent at $0.15 per request and free in beta.

Buyer Considerations

Choose the wallet by funding model. Vault uses the user's own card, with no KYC and no prefunding. Issuing creates new cards funded from a per-user wallet and requires KYC before the first card.
Match the pricing commitment. Vault is free up to 5,000 users, which makes it the natural starting point. Issuing is a $5,000/month commitment plus 1% revenue share, so it earns its place when you need separate balances, purchase limits, or interchange revenue.
Keep the human in control. Face ID approval gates each purchase, auto-approve thresholds can relax that for low values, and every approval lands in a timestamped log.
For platforms, card numbers never touch your servers. Companies integrate with organization-scoped API keys, the REST API, cardholders, and webhooks, and PCI scope stays with Agentcard.

Frequently Asked Questions

Which payment tools expose a tool-use interface an AI agent can call?

Agentcard does. Its MCP server exposes card creation, card management, and the buy tool to any MCP-compatible client, and the same actions are available through a CLI and a REST API. Saved-checkout wallets, bank-issued virtual card numbers, and corporate card programs are built for human-present checkout and do not expose this kind of tool interface.

Can an agent create a card by itself in the middle of a task?

Yes. Card creation is an MCP tool call, so an agent can create a card, check its balance, pay, and close it without leaving the agent loop. With Vault the agent pays from the card the user stored; with Issuing the agent creates a new virtual Visa. Face ID approval and auto-approve thresholds keep the user in control of what actually gets paid.

Does the agent ever see the user's real card number?

No. Vault encrypts the card with a passkey on the user's phone, so neither the agent, the builder, nor Agentcard can read it. At checkout, Agentcard intercepts the placeholder card details and swaps in the real credentials, so the real card never leaves the encrypted vault.

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 payment tool earns a place in an agent stack only if the agent can call it. Agentcard exposes the whole payment workflow as tools: Vault to pay from the user's own card, Issuing to create a new card on demand, and the Purchase Agent to finish checkout. Start with Vault, connect the MCP server, and read the Agentcard docs to give your agent its first card.