Table of contents
Is there an official Stripe MCP server? Yes - Stripe hosts one directly at mcp.stripe.com, and separately publishes an open-source local server through its Agent Toolkit. This information is correct as of September 2026.
If you’re still weighing up MCP against calling a tool’s API directly, or want the fundamentals of the protocol first, our MCP introduction is the place to start. Here’s what Stripe’s MCP server actually exposes, how the hosted and local options differ, how to set up the one most marketing and RevOps teams want, and a change landing before the end of October that affects anyone using API-key auth.
Two ways to run the Stripe MCP server
The hosted MCP server (recommended for most teams)
Stripe’s hosted MCP server runs at https://mcp.stripe.com. It’s Stripe-managed, supports OAuth so your agent can act on your behalf without you handing over a raw API key, and also accepts an Agent API key as a bearer token for autonomous clients that can’t complete an OAuth flow. Stripe recommends installing it through the Stripe CLI’s agent plugin (stripe agent setup), which configures the MCP server, installs Stripe’s supporting “skills,” and keeps both up to date automatically. Manual setup instructions also exist for Cursor, Claude Code, Claude Desktop, Codex, ChatGPT and VS Code individually, each with a one-click install link or a short JSON snippet.
The local Agent Toolkit server
Stripe also ships an MIT-licensed, open-source local server as part of its Agent Toolkit, the same repo behind its LangChain, CrewAI, OpenAI Agent SDK and Vercel AI SDK integrations. Run it with:
npx -y @stripe/mcp --tools=all --api-key=YOUR_STRIPE_SECRET_KEYInstead of the hosted server’s nine broad tools, the local server lets you enable individual actions per resource, for example customers.create, paymentLinks.create or products.create, so a given deployment only ever has access to the specific writes you’ve explicitly turned on.
Setting up the hosted Stripe MCP server
- Install via the plugin (recommended). Run
npm install -g @stripe/cli@latestthenstripe agent setup. This detects which agents you use and configures the Stripe MCP server and skills for each one automatically. - Or add it manually. Point your client at
https://mcp.stripe.com. Cursor, Claude Code, Codex and VS Code all have one-click install links or a short JSON snippet; Claude Desktop and ChatGPT use their own connector or plugin directories instead. - Authenticate. Use OAuth for any interactive client, granting access to specific live or sandbox accounts with their own permissions. For autonomous or headless clients, create a restricted Agent API key instead and pass it as a bearer token.
- Manage access centrally. Admins can turn MCP access on or off for the whole team, separately for live mode and sandboxes, from the Dashboard’s MCP and CLI access settings, and revoke individual sessions from Team and security at any time.
The freshness detail that matters right now: from 31 October 2026, the Stripe MCP server stops accepting full-access secret keys or restricted keys that don’t carry the Agent tag. Any integration still authenticating with an old-style key needs to move to a proper Agent Key or OAuth before then, or it starts getting a 401 response with an OAuth discovery challenge.
What tools does the Stripe MCP server expose
| Resource | Tool | Description |
|---|---|---|
| API tools | stripe_api_search | Search for Stripe API methods by keyword |
| API tools | stripe_api_details | Get parameter details for a specific Stripe API method |
| API tools | stripe_api_read | Read data with any Stripe API GET method |
| API tools | stripe_api_write | Write data with any POST, PATCH, PUT or DELETE method |
| Account | get_stripe_account_info | Retrieve account details |
| Analytics (preview) | stripe_analytics | Query metrics, analyse subscriptions and billing, run financial reports |
| Treasury (public preview) | get_balance_summary | Interactive balance summary across the Stripe balance and Treasury accounts |
| Other | search_stripe_documentation | Search Stripe’s documentation and support articles |
| Other | stripe_implementation_planner | Guide through Stripe products for a given integration goal |
| Other | send_stripe_feedback | Share feedback about Stripe’s products |
The two generic tools, stripe_api_read and stripe_api_write, are doing most of the work: between them they cover roughly 150 underlying API methods across customers, charges, refunds, payment intents, checkout sessions, invoices, subscriptions, coupons, promotion codes, products, prices, payment links, disputes, payouts, tax and more, without every single method needing its own dedicated tool and bloating the context window.
Marketing and GTM use cases
Once connected, the practical value for a marketing or RevOps team looks like:
- Payment link and campaign performance - “How much revenue came through the payment link we sent in last week’s email campaign?” answered from live checkout session and payment link data via
stripe_api_read, instead of exporting a CSV from the Dashboard. - Subscription and churn queries - asking an agent to list subscriptions cancelled in the last 30 days for a specific price, or check a named customer’s subscription status, to sanity-check a retention campaign’s read on churn.
- Coupon and promotion creation for campaigns - “Create a 20% coupon for the Q4 launch and a promotion code called LAUNCH20” via
stripe_api_write, straight from the same conversation where the campaign is planned. - Revenue sense-checks alongside traffic metrics - pulling balance, payout or invoice data next to a campaign’s session numbers, so a “did this actually move revenue” question doesn’t need a separate finance export.
Stripe MCP options compared
| Hosted MCP server | Local Agent Toolkit server | |
|---|---|---|
| Status | Official, Stripe-hosted, recommended | Official, open source |
| Hosting | Remote, at mcp.stripe.com | Local, runs via npx @stripe/mcp on your machine or infrastructure |
| Auth | OAuth (recommended), or an Agent API key as a bearer token | Stripe secret or restricted API key; Agent Keys required from 31 October 2026 |
| Setup | stripe agent setup (CLI plugin, auto-configures and keeps skills updated) or manual per-client config | npm install -g @stripe/cli then npx -y @stripe/mcp --tools=... --api-key=... |
| Tool surface | 9 tools: generic read/write covering ~150 API methods, plus docs search, analytics and account info | Granular per-resource actions enabled individually (e.g. customers.create, paymentLinks.create) |
| License | N/A (hosted service) | MIT, github.com/stripe/agent-toolkit |
| Best for | Most marketing and RevOps teams, connecting via Cursor, Claude, ChatGPT, Codex or VS Code | Custom or headless integrations needing tightly scoped, resource-level tool access |
See the full MCP servers list for how Stripe’s server fits alongside the rest of a marketing team’s stack.
Can it move money? Confirmation and access control
The honest answer is yes, and that’s exactly why Stripe puts a human checkpoint in front of the actions that matter. Stripe requires manual confirmation before certain stripe_api_write actions complete, refunds and outbound payments among them: your agent gets a URL to click, you review the request in the Dashboard, and only then does Stripe hand back an approval token for the agent to retry. Unapproved requests expire after 24 hours.
That still leaves reads and lower-risk writes, like creating a coupon or drafting an invoice, able to run without a click, so scope access properly: use a restricted Agent API key with only the permissions your agent needs rather than a full-access secret key, and apply the same prompt-injection caution here that you would with any tool that has write access into a live system, especially when chaining the Stripe MCP server with others in the same agent session.
About Growth Method
Stripe’s MCP server puts your revenue data one conversation away; Growth Method puts the campaign that revenue is meant to prove out in the same place. We’re the agentic marketing platform for B2B teams: a campaign’s hypothesis, plan and results live together, so plan, launch and analyse becomes one continuous loop instead of a plan in one tool and the proof in another. Connect Stripe alongside GA4, PostHog, Search Console and the rest of your stack, and both your team and its AI agents can check whether a campaign actually moved revenue, not just traffic, from the same live data these MCP servers expose.
Get started to connect your stack and tie your next campaign to the revenue it’s meant to move.
Frequently asked questions
Is there an official Stripe MCP server?
Yes. Stripe hosts an official remote MCP server at mcp.stripe.com, authenticated via OAuth or an Agent API key, and also publishes an open-source local server through its Agent Toolkit (MIT-licensed, on GitHub as stripe/agent-toolkit), run with npx @stripe/mcp. Most teams should start with the hosted server; the local option suits custom or headless integrations.
What’s the difference between the hosted Stripe MCP server and the local Agent Toolkit?
The hosted server at mcp.stripe.com is Stripe-managed, supports OAuth, and exposes nine tools including generic read and write methods that cover roughly 150 API endpoints. The local Agent Toolkit server runs on your own machine or infrastructure via npx, authenticates with a Stripe secret or restricted key, and lets you enable individual granular actions, like customers.create or paymentLinks.create, one by one. Use the hosted server unless you specifically need local hosting or fine-grained per-action scoping.
Is the Stripe MCP server free to use?
Yes, connecting either server is free. You’re not charged for MCP access itself; you pay only Stripe’s normal processing fees on any payments the tools touch, the same as if you’d made the API call or Dashboard change yourself.
Can the Stripe MCP server move money, and is it safe to connect?
It can, which is why Stripe requires human confirmation before certain stripe_api_write actions, such as refunds and outbound payments, complete. Your agent gets a URL to click and approve before the action goes through, and unapproved requests expire after 24 hours. Scope access with a restricted Agent API key rather than a full-access secret key, and apply the same prompt-injection caution you’d use with any tool that has write access to your systems.