Stripe AI
Connect AI products to Stripe billing, usage metering, MCP, and development skills.
- Source repo
- stripe/ai
- Stars
- ★ 2k
- Last updated
- today
- License
- MIT
- Primary language
- TypeScript
- FA score
- 50/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code · OpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A TypeScript team using Vercel AI SDK can connect its AI product to Stripe's billing infrastructure through
@stripe/ai-sdk. - Not a fit if
- Teams avoiding Stripe billing lock-in; switching providers means replacing the integration.
- Teams wanting a ready-made agent app; this repo only ships SDKs and plugins.
- Source review
- 50/100 · Major gaps 2 safety controls not found
What does this agent do, and when should you use it?
Stripe AI is a collection of SDKs and integrations for developers building AI products and businesses, rather than a standalone runnable agent. It includes @stripe/ai-sdk, which connects Stripe billing infrastructure to Vercel's ai and @ai-sdk libraries. It also provides @stripe/token-meter for integrations using the native OpenAI, Anthropic, and Google Gemini SDKs without a framework dependency. Stripe hosts an OAuth-enabled remote MCP server at https://mcp.stripe.com and links this interface to autonomous-agent development. Official plugins are documented for Claude Code, Codex, Cursor, and Grok Build, while Agent Skills can also be installed manually with npx. The documented boundary is the Stripe integration and developer-tooling layer; the source does not present a complete hosted agent, application runtime, or end-user interface.
@stripe/ai-sdk integrates Stripe's billing infrastructure with Vercel's ai and @ai-sdk libraries. @stripe/token-meter supplies metering integration for the native OpenAI, Anthropic, and Google Gemini SDKs and explicitly avoids a framework dependency. MCP clients can connect to https://mcp.stripe.com and authenticate through OAuth to access Stripe; the same MCP interface is presented as a foundation for autonomous agents. Official Stripe plugins add Agent Skills and other agent tools to Claude Code, Codex, Cursor, and Grok Build, while the manual path installs skills from https://docs.stripe.com. The source does not specify SDK classes, function calls, metering event schemas, or generated Stripe objects, so a more detailed execution flow cannot be verified.
- A TypeScript team using Vercel AI SDK can connect its AI product to Stripe's billing infrastructure through
@stripe/ai-sdk. - A developer calling the native OpenAI SDK can add AI usage metering without adopting an agent framework.
- A SaaS team using the native Anthropic or Google Gemini SDK can evaluate
@stripe/token-meterfor a shared Stripe-oriented metering path. - An engineer building Stripe functionality in Claude Code or Codex can install the official Stripe plugin for skills and additional agent tools.
- A team creating an autonomous agent can use Stripe's OAuth-enabled remote MCP endpoint as its Stripe access interface.
- A developer whose harness lacks the official plugin can install Stripe's skills manually and manage updates with
npx.
How do you install or deploy this agent?
For Claude Code, run claude plugin install stripe@claude-plugins-official. For Codex, run codex plugin add stripe@openai-curated. In Cursor, run /add-plugin stripe, or install through the Cursor marketplace. For Grok Build, run grok plugin install stripe --trust. To install the skills manually, run npx skills add https://docs.stripe.com; manual installations do not auto-update, so use npx skills update -y for later updates. The source does not provide package-manager installation commands for @stripe/ai-sdk or @stripe/token-meter, minimum runtime versions, or the required Stripe credential configuration.
How do you use this agent?
From an MCP-capable client, connect to https://mcp.stripe.com; access is secured with OAuth, although the source provides neither a complete client configuration nor a first-call command. In a supported coding harness, install the corresponding official plugin to make Stripe Agent Skills and additional tools available. For a manual skills setup, run npx skills add https://docs.stripe.com and maintain it with npx skills update -y. Choose @stripe/ai-sdk for Vercel AI SDK projects or @stripe/token-meter for native OpenAI, Anthropic, or Google Gemini SDK integrations. Imports, initialization arguments, credential environment variables, and a first executable API invocation are not documented in the supplied material.
What are this agent's strengths and limitations?
- It covers both Vercel AI SDK and native OpenAI, Anthropic, and Google Gemini SDK integration paths.
@stripe/token-meterhas no framework dependency, which suits applications that want metering without adopting an agent framework.- Stripe operates an OAuth-enabled remote MCP endpoint, removing the need to deploy that endpoint yourself.
- Claude Code, Codex, Cursor, and Grok Build each have an explicit official plugin installation path.
- The repository is licensed under MIT.
- This is not a complete runnable agent; adopters must still supply the application, model-call flow, and business logic.
- The central functionality depends on Stripe's billing infrastructure, so moving to another billing provider may require replacement integration work.
- The supplied source omits SDK installation commands, API examples, runtime versions, and credential setup, preventing a verified first SDK invocation.
- The remote MCP route depends on network access and OAuth, with no documented offline or self-hosted alternative.
- Manually installed skills do not update automatically and require
npx skills update -y. - The source does not document metering accuracy, retries, idempotency, failure recovery, or provider-specific limitations.
How does this agent compare with similar options?
@stripe/ai-sdk targets Vercel's ai and @ai-sdk libraries, while @stripe/token-meter works with the native OpenAI, Anthropic, and Google Gemini SDKs without a framework dependency. The former fits an existing Vercel AI SDK application; the latter fits teams that want to retain native provider SDK calls.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Stripe AI This agent | 50 · Major gaps | Library / SDKFree + model costs | ★ 2k | today | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
| AgentKit (Coinbase Developer Platform's AI Agent Crypto Wallet Toolkit) | 44 · Major gaps | CLIFree + model costs | ★ 1.3k | 4d ago | TypeScript | OpenAI API |
| mcp-agent | 53 · Major gaps | FrameworkFree + model costs | ★ 8.6k | 8mo ago | Python | Claude.ai · OpenAI API · Claude API |
| Crush Terminal Coding Assistant | 43 · Major gaps | CLIFree + model costs | ★ 29k | today | Go | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
The README states that the remote MCP uses OAuth, the test concretely shows customer identifiers, model names, and token usage being sent to Stripe, and CI defaults repository contents access to read-only. This makes the principal data flow and some permissions visible. However, there is no complete runtime permission inventory, user-confirmation mechanism, credential-lifecycle guidance, or revocation and rollback procedure. Meter events are sent asynchronously, with no shown handling for failure, retries, or duplicate billing. The security policy supplies a clear vulnerability-reporting route and frozen-lockfile installation provides baseline supply-chain control, but dependency scanning, vulnerability policy, and comprehensive pinning are not evidenced. Stripe naming, copyright, and centralized-sync notices support attribution, although publisher identity remains unknown under the supplied registry rule and earns no inferred trust.
The Stripe metering description is consistent with the Anthropic wrapper tests, while CI covers builds, formatting, linting, and tests for TypeScript, MCP, and Python. Fixed pnpm, Node, and Python versions plus frozen lockfiles support ordinary dependency availability, although operation still relies on external registries, hosted MCP, plugin marketplaces, and third-party model SDKs. Validation scripts expose some progress and exit-code diagnostics, but server waits lack timeouts and the fire-and-forget metering path has no demonstrated error reporting or recovery semantics, preventing full marks.
The material identifies AI products, token metering, MCP, autonomous agents, framework integration, and framework-independent SDK use as target scenarios. Installation paths for Claude Code, Codex, Cursor, and Grok, together with TypeScript, Python, and MCP CI coverage, show useful environment fit. Deductions reflect the absence of a complete capability boundary, support matrix, prerequisites, and unsuitable-use cases. Skill triggering is described only broadly, without precise activation, exclusion, or conflict rules.
The README has a clear structure around SDKs, MCP, skills, per-harness installation, and licensing; package names and repository paths are stable within the supplied evidence. Manual installation explicitly warns that updates are not automatic. The complete MIT text justifies full license credit. Examples are largely installation commands and isolated test usage, with no FAQ, full workflow, or troubleshooting guide. Known limitations are restricted to the manual-update warning, and no versioning policy, release notes, or changelog is supplied. SECURITY.md, the centralized-sync workflow, and a named maintenance team establish useful maintenance paths, while publisher identity remains unverified as instructed.
The SDKs, framework-independent token metering, hosted MCP, and multi-harness distribution form a usable Stripe AI integration entry point. Tests demonstrate the concrete output of normalized model identifiers and separate input/output meter events, supporting practical and marginal value over bespoke integration. The evidence does not quantify end-to-end outcomes, performance, accuracy, operating cost, billing risk, or migration effort, so cost-benefit support remains thin.
Core claims trace to named packages, commands, CI workflows, and specific metering tests, with useful corroboration across the README, tests, license, security policy, and sync workflow. The score is reduced because the supplied evidence is only a repository slice and omits core implementation, lockfiles, skill contents, and release history; the “one-stop shop” claim is also not quantitatively supported. Descriptions, commands, and test expectations are generally distinguishable, but there is no explicit discipline for labeling facts, inferences, and unverified claims.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Metering events contain Stripe customer identifiers, model names, and token counts; verify data minimization, retention, logging, and privacy requirements before deployment.
- The test describes metering submission as fire-and-forget, but the evidence does not show failure handling, retries, idempotency, or duplicate-billing protection.
- The hosted MCP, plugin marketplaces, and third-party model SDKs are external dependencies; verify current permissions, availability, and update provenance before adoption.
- Manually installed skills do not update automatically, and direct changes to synchronized directories may be overwritten.
- This assessment is based only on the supplied static files; no code, tests, or network interactions were executed.
FAQ
Is this a deployable autonomous agent?
Do I have to use Vercel AI SDK?
@stripe/ai-sdk is intended for Vercel AI SDK, while @stripe/token-meter supports native OpenAI, Anthropic, and Google Gemini SDK integrations without a framework dependency.What permissions does the MCP server require?
What will this integration cost?
How are plugins and manually installed skills updated?
npx skills update -y.