Your AI agent can move money in Stripe - here's how to stay in control
Source
stripe.dev
Author
Daniel Wood
Date
Why it matters
Explains the two models for giving an agent Stripe access and shows that live refunds/payouts always require a human confirmation step even with auto-approve clients disabled, a concrete safety guarantee builders can rely on.
Key takeaways · AI-distilled
Stripe describes two ways to give an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → access: OAuth through its MCPThe Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.Full definition → server, where the agent acts as your user with per-account read or write scopes, or an agent-tagged API key that gives a headless agentAn agent running with no one watching — in CI, on a schedule, or triggered by an event — with its output reviewed after the fact.Full definition → its own long-lived identity.
Agent-tagged keys fall under Approval rules: a condition such as Refund is created with Amount > $500 routes the action to designated reviewers, and by default no actor, human or agent, can approve its own request.
Stripe ships default rules scoped to agent-tagged keys covering refunds, payouts and outbound transfers, so a baseline of review applies as soon as the key is created; teams then tune reviewers, conditions, and whether to require approval or block.
A rule-gated call returns an approval_required error instead of completing, and Stripe files an approval request the agent can annotate with a reason. Approved requests are attempted automatically; requests not approved within 14 days expire.
Agents can follow outcomes through v2.core.approval_request webhooks (created, approved, rejected, canceled, expired, succeeded, failed), so a support bot learns whether a refund it requested was actually completed.
Terms in this piece · Glossary
MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
headless agent — An agent running with no one watching — in CI, on a schedule, or triggered by an event — with its output reviewed after the fact.
