stripe/ai
Project Overview
Stripe AI
Official Stripe SDKs, MCP servers, and agent toolkits for integrating Stripe payments and billing with LLMs and AI agents.
True Intent (AI Assessment)
Provide official, multi-language SDKs and MCP tooling that let AI agents and LLM apps use Stripe for payments, billing, and usage-based token metering.
README Accuracy
Notes
First-party Stripe repository. Package names, authorship (support@stripe.com), CODEOWNERS, and MIT license are all consistent. No obfuscation, network exfiltration, or suspicious install scripts observed.
This is Stripe's official monorepo for AI integration tooling. It bundles together several published packages (@stripe/ai-sdk, @stripe/token-meter, @stripe/agent-toolkit, @stripe/mcp, and a Python stripe-agent-toolkit) that let developers connect Stripe's payment and billing APIs to LLM-based applications and autonomous agents. It also ships agent 'skills' and plugin manifests for Claude Code, Codex, Cursor, and Grok, plus a set of benchmark environments for evaluating how well AI agents can perform Stripe integration tasks.
The problem it solves is the gap between LLMs/agents and real-world commerce operations. Developers building AI apps need standardized ways to (a) let agents call Stripe APIs safely (via MCP or framework-specific tool wrappers for LangChain, Vercel AI SDK, OpenAI Agents, CrewAI, Strands), and (b) meter and bill for LLM token usage through Stripe's billing infrastructure. This repo provides both directions in one place.
Users are developers building AI products that either transact through Stripe or want to charge customers based on AI token consumption. They install the appropriate package for their stack — TypeScript devs use @stripe/agent-toolkit or @stripe/ai-sdk, Python devs use stripe-agent-toolkit, and agent-harness users install the Stripe plugin into Claude Code, Cursor, Codex, or Grok. The MCP server at mcp.stripe.com can also be connected directly via OAuth.
Under the hood, the toolkits wrap the official Stripe SDK (stripe ^17.5 for Node, stripe for Python) and expose its operations as tool definitions in each framework's format (Zod schemas + zod-to-json-schema for TS, Pydantic for Python). The token-meter package intercepts calls to OpenAI/Anthropic/Google SDKs to report usage to Stripe Meters. The MCP tool package uses @modelcontextprotocol/sdk to expose Stripe capabilities to any MCP client. Benchmarks are Dockerized Ruby/Python/JS checkout environments with graders that score an agent's completed integration.
No red flags: MIT licensed, code owned by Stripe, dependencies are mainstream and pinned, CI includes CodeQL and dependabot, and pnpm overrides patch known-vulnerable transitive deps (form-data, brace-expansion, tmp, etc.). Stated purpose matches observed code.
| Languages | TypeScript, Python, Ruby, JavaScript, Java, C# |
| Runtime | Node.js >=18, Python >=3.11 |
| Framework | Model Context Protocol (MCP), Vercel AI SDK, LangChain, OpenAI Agents, CrewAI |
| Database | None detected |
| Package Manager | pnpm (Node), pip/setuptools (Python), Bundler (Ruby benchmarks) |
| Key Dependencies | stripe, @modelcontextprotocol/sdk, @ai-sdk/provider, zod, openai, @anthropic-ai/sdk, @google/generative-ai, pydantic |
| Build Tool | tsup / tsc for TS packages, setuptools for Python |
| Test Framework | Jest (TS), pytest (Python), RSpec (Ruby benchmarks) |