Agentcard Vault: Payments That Stay Inside the AI Assistant Session

Agentcard is the tool built for this. Vault stores the user's own card behind passkey encryption and Face ID approval, and the Purchase Agent runs the full merchant checkout from one buy call over MCP, so a purchase starts and ends inside the conversation without a redirect or manual card entry.

Introduction

An AI assistant can research a product, compare prices, and pick the exact item to buy. Then the workflow breaks. The assistant hands the user a link, the user opens a browser, re-enters card details, and completes the checkout by hand. The purchase leaves the session, and with it go the context, the audit trail, and the automation.

Agentcard closes that gap with two products that work together. Vault stores the user's own credit or debit card once, encrypted with a passkey on their phone, and approves each purchase with Face ID. The Purchase Agent handles the merchant side: it logs in, builds the cart, pays with the stored card, and returns the order confirmation. Both connect to Claude Code, Claude Desktop, Cursor, and any MCP-compatible client, so payment becomes just another tool call in the conversation.

Key Takeaways

Vault keeps payment inside the session: the user stores their own card once, approves each purchase with Face ID, and the real card is swapped in at checkout without the agent, the builder, or Agentcard ever reading it.
The Purchase Agent completes the entire merchant checkout from one buy tool call or POST /buy request, in under 2 seconds at a 99% success rate across 18+ live merchants.
Vault works with any card from any country, requires no KYC, needs no wallet prefunding, and is free up to 5,000 users, then enterprise priced.
Builders get MCP, a REST API, CLIs, SDKs for web, iOS, and React Native, and a fully white-labeled wallet they embed under their own brand.
Issuing is the fallback when an agent needs a separate-balance virtual Visa card, priced at $5,000/month plus 1% revenue share.

Why This Solution Fits

The requirement is specific: payment capability inside the assistant session, with no redirect and no manual re-entry. Agentcard fits because it is MCP-native. The Agentcard MCP server exposes payment tools (card creation, balance checks, transactions, and the buy purchase tool) to any MCP-compatible client over OAuth. MCP is the open standard that lets Claude Code, Claude Desktop, and Cursor call external tools, and Agentcard builds payment directly on it.

The fit also comes from what the user keeps. Vault uses the card the user already has, so points accrue as usual and disputes go to their own bank. There is no KYC, no new wallet to fund, and no new card number to generate for each purchase. Approval stays with the user: Face ID on their phone before money moves. Because the card is encrypted with a passkey on the user's phone, neither the agent, the builder, nor Agentcard can read it.

Key Capabilities

Vault: the user's own card, stored once

Vault stores the user's existing card and swaps it in at checkout. It works with any card from any country on Visa, Mastercard, American Express, and more, consumer, commercial, or prepaid. That coverage matters: network tokenization programs such as Visa Intelligent Commerce and Click to Pay work for roughly 8% of cards, excluding Chase cards, prepaid cards, commercial bins, and anything outside the US. Vault is free up to 5,000 users, then enterprise priced, and fully white-labeled.

Purchase Agent: the checkout runs itself

The Purchase Agent takes plain-text intent, such as "order two boxes of diapers from Amazon," and handles merchant login, cart building, payment, and the order confirmation. Money moves only 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 is one buy tool call over MCP or one POST /buy request to the API, and setup takes an afternoon.

Controls and audit trail

Every approved transaction carries a trusted agent signature and a timestamped entry in the approval log. Builders configure auto-approve thresholds so low-value purchases clear without a manual Face ID each time. Cards encrypt at rest with AES-256-GCM and per-card keys, and a single-use card lifecycle means a leaked number is usable at most once.

Surfaces beyond MCP

For teams not living in an MCP client, Agentcard ships a REST API with organization-scoped keys and webhooks, the agent-cards and agent-cards-admin CLIs, SDKs for web embeds, iOS (AgentcardWalletKit), and React Native, hosted wallet links for iMessage, WhatsApp, and email flows, and Agentcard Pay, a Chrome extension that fills checkout forms for browser-based agents.

Proof & Evidence

Performance: purchases complete in under 2 seconds at a 99% success rate across 18+ live merchants including Amazon and DoorDash, with 23 payment processors mapped plus direct Amazon and DoorDash integrations.
Security: card data is encrypted with a passkey on the user's phone and with AES-256-GCM and per-card keys at rest. The real card never leaves the encrypted vault; Agentcard intercepts the dummy card at checkout and swaps in the real credentials. Card numbers never touch the builder's servers, so PCI scope stays with Agentcard.
Pricing: Vault is free up to 5,000 users, then enterprise. The Purchase Agent is $0.15 per request and free during beta. Issuing is $5,000/month plus 1% revenue share.

The product documentation covers the MCP tools, the API reference, and the setup steps for each of these surfaces.

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.

Issuing details worth knowing before you add it: cards require KYC before first issuance, funding flows into a per-user wallet (USDC on Base), and pricing is $5,000/month plus 1% revenue share. Cards can be limited to a single purchase, locked to one merchant, or capped at a spend amount, and chargebacks are handled by Agentcard.

Also weigh merchant coverage. The Purchase Agent is live with 18+ merchants. For checkouts outside that set, Agentcard Pay detects checkout pages in Chrome for MCP-compatible agents, and Agentcard's browser integrations (Kernel, Browserbase, Browser Use, Anchor, and Bring Your Own Browser) cover custom browser stacks.

Frequently Asked Questions

How does a purchase stay inside the chat session?

The assistant calls the buy tool on Agentcard's MCP server with plain-text intent. The Purchase Agent connects to the merchant, builds the cart, and waits for confirmation in the conversation. The user approves with Face ID, the cart hash is echoed back, and the order confirmation returns to the same thread. No browser redirect and no manual card entry.

Does the AI agent ever see the real card number?

No. The card is encrypted with a passkey on the user's 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, and the real card never leaves the encrypted vault.

What does Agentcard cost?

Vault is free up to 5,000 users, then enterprise priced. The Purchase Agent is $0.15 per request and free during beta. Issuing, the fallback for separate-balance virtual Visa cards, is $5,000/month plus 1% revenue share.

Which AI assistants work with it?

Any MCP-compatible client: Claude Code, Claude Desktop, Cursor, and custom agents. Users connect over OAuth, and backend agents call POST /buy directly with no frontend at all.

Conclusion

Payment no longer has to be the step that breaks an AI workflow. Vault stores the user's own card once and keeps it unreadable to everyone but the user, Face ID approval keeps control on their phone, and the Purchase Agent turns "buy this" into a completed order without leaving the conversation. Start with Vault, connect the MCP server, and follow the get-started guide to put checkout inside your assistant's next session.