What it does

The server exposes Stripe payment and billing APIs to LLM agents through the Model Context Protocol. It can run either locally via npx @stripe/mcp or as a remote OAuth-authenticated server at https://mcp.stripe.com. Tool availability is controlled via Stripe's Restricted API Keys, allowing fine-grained permission scoping. Works with agent frameworks including OpenAI's Agent SDK, LangChain, CrewAI, and Vercel's AI SDK.

Who it's for

Builders implementing AI agents that need to perform billing, payment, or customer operations on Stripe—engineers wiring Claude or other LLMs into Stripe accounts for autonomous financial workflows, subscription management, or payment processing.

Common use cases

Automate customer invoice and payment reconciliation through an agentic interface
Generate refunds or adjust subscriptions via natural language commands to an AI agent
Query Stripe customer and transaction data to provide billing context to LLMs
Build AI-powered support agents that can modify account status or process payments
Create autonomous billing workflows (like refund decisions, retry logic) controlled by LLMs

Setup pitfalls

API key security: The scanner detected one hardcoded or exposed secret; never commit STRIPE_SECRET_KEY or STRIPE_RESTRICTED_KEY to version control. Use environment variables or a secrets manager.
Restricted API Key (RAK) permissions: Tool availability depends entirely on the RAK's configured permissions. Test scoping at https://dashboard.stripe.com/apikeys before deploying agents.
Remote vs. local trade-off: The remote server at mcp.stripe.com requires OAuth setup; the local npx server needs your secret key passed at startup. Both require secure credential handling.
Filesystem access: The server reads and writes files; ensure it runs with appropriate sandboxing and restricted working directory permissions.