Stripe builds the two-way bridge between AI agents and money

What it does

This repo ships three SDKs that wire Stripe into the modern AI stack. @stripe/agent-toolkit exposes Stripe API methods as function-calling tools for LangChain, CrewAI, and OpenAI’s Agent SDK. On the flip side, @stripe/ai-sdk and @stripe/token-meter let you meter and bill customers for LLM usage through Vercel’s AI SDK or directly via OpenAI, Anthropic, and Gemini native clients.

The interesting bit

Stripe is treating agent access as a first-class security concern. It hosts a remote MCP server with OAuth and ships a local MCP wrapper, while tool permissions are gated by Restricted API Keys rather than all-or-nothing secrets. The libraries are built directly on Stripe’s existing Python and Node SDKs, so this is well-structured glue code with granular scoping.

Key highlights

• Adapters for LangChain, CrewAI, OpenAI Agent SDK, and Vercel’s AI SDK in Python and TypeScript
• @stripe/token-meter handles usage billing without tying you to a specific framework
• Remote MCP server at mcp.stripe.com plus local MCP exposure through the toolkit
• Tool access controlled by Restricted API Keys, not blanket secret-key privileges
• Context passing for Stripe Connect connected accounts

Caveats

• The agent toolkit explicitly does not cover the entire Stripe API
• A migration guide from v0.8.x suggests the TypeScript interface recently shifted
• The repo is a loose collection of packages rather than a single unified framework

Verdict

Grab it if your agents need to charge cards or refund customers, or if you need to bill for tokens. If your AI work never touches money, you can safely ignore it.