AgentPay Docs
Welcome
AgentPay gives AI agents the ability to transact within policy guardrails. You save a payment method once; your agent creates virtual cards against it; each card has a spending limit, metadata, and optional human approval. No multi-tenant Stripe onboarding for your users; no custody of funds.
5-minute quickstart
Sign up, complete KYB, install the MCP server, issue your first card via curl.
Read →
API reference
Every endpoint with curl, TypeScript, and Python examples.
Read →
MCP server
Copy-paste config for Claude Desktop, Cursor, and Windsurf.
Read →
Concepts
Cardholders, cards, holds, policies, approvals. The funding model explained honestly.
Read →
Mental model, in one sentence
You keep your payment method; each card request places a temporary hold on it for the card's spend limit, mints a single-use Stripe Issuing virtual card, and releases or captures the hold based on whether the card is actually used.
What's different about AgentPay
- Policy as a first-class primitive. Default + per-card override. Driven via API, not buried in dashboard settings. Policy cookbook.
- Silent auth-holds. No Stripe Checkout per card by default. The user saves a payment method once and agents mint cards without blocking the flow.
- Per-user KYB cardholders. Real compliance, not shared-address hacks. Your users' AVS checks pass at merchants.
- Structured errors the agent can relay. When setup is incomplete, we return a typed error with a
setup_urlso the agent tells the user exactly what to do.
What's not in v0
- Multi-use cards (SaaS renewals / recurring). Planned for v1 — the #1 user request.
- Wallet / pre-funded balance. Requires Treasury or BaaS; planned for v2.
- MCC merchant restrictions. Planned for v1 alongside multi-use.
- CLI, Chrome extension. Not planned — we invest in MCP quality instead.