Stripe MCP Server

The Stripe MCP Server (currently in public preview per Stripe's documentation) implements the Model Context Protocol and gives AI agents access to the Stripe API: search Stripe API methods by keyword and look up their parameters, run arbitrary read (GET) and write (POST/PATCH/PUT/DELETE) API calls, retrieve account information, and search Stripe documentation and support articles. Additional tools cover analytics (querying metrics, time-series charts, Sigma queries), Treasury balances, reports, and an implementation assistant that guides you through building a Stripe integration.

Setup

Stripe recommends setting it up via an agent plug-in that configures the MCP server automatically and keeps it up to date:

npm install -g @stripe/cli@latest
stripe agent setup

Alternatively, register the server manually, e.g. in Claude Code with claude mcp add --transport http stripe https://mcp.stripe.com, or in Cursor via an mcpServers entry with "url": "https://mcp.stripe.com". The connection uses OAuth: on the first call, Stripe's consent page opens to confirm access to your own Stripe account.

Stripe explicitly recommends enabling human confirmation for tool calls and being cautious of prompt injection risks when using multiple MCP servers at once.

Why there's no self-hosting option

Unlike many other providers, Stripe does not offer a separate, open-source server component for self-hosting. The reason lies in the nature of the data being processed: per the provider, payment, customer, and revenue information should run exclusively through infrastructure controlled and vetted by Stripe itself, rather than through arbitrarily configured third-party servers. For users, this means less operational overhead but also less control over the exact server location and runtime environment compared to open-source MCP servers from other providers.

Typical use cases

The server is particularly suited for support and accounting tasks where an agent needs to look up payment status, customer data, or subscriptions without anyone manually switching between the Stripe Dashboard and a chat. It's also useful for building new Stripe integrations, where the implementation assistant walks you step by step through API methods and parameters. For automated, recurring workflows such as monthly revenue reporting, the analytics and reporting tools are a good fit, but should always run with restricted API keys instead of full account access given the write capability involved.

Frequently asked questions

Can I self-host the Stripe MCP Server? No, it runs exclusively as a Stripe-hosted service; there's no open-source alternative for running it yourself. Does the server work without the Stripe CLI? Yes, manual setup via the client's MCP configuration works just as well; the CLI is only the recommended, automated path. Is the server suitable for live payment data? Yes, but Stripe explicitly recommends human confirmation for write actions and separate controls for live and sandbox environments.

A Stripe account plus an MCP client that supports remote servers over HTTP and OAuth. The recommended plug-in setup additionally requires Node.js/npm for the Stripe CLI.

Recommended: install the Stripe CLI and run the setup assistant, which automatically configures the detected agent:

npm install -g @stripe/cli@latest
stripe agent setup

Manually in Claude Code:

claude mcp add --transport http stripe https://mcp.stripe.com

Manually in Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "stripe": { "url": "https://mcp.stripe.com" }
  }
}

On the first call, the client opens Stripe's OAuth consent page; confirm access to your own account there.

claude mcp add --transport http stripe https://mcp.stripe.com

OAuth authorization per the MCP specification, triggered on first connection. Authorized clients and their sessions can be managed and individually revoked in the Stripe Dashboard under user settings.

By default the tools operate on the Stripe account tied to your own OAuth session or API key. API access via stripe_api_read/stripe_api_write covers most of the Stripe API for both reading and writing. Acting on behalf of connected accounts (Connect platforms) requires a restricted API key instead of OAuth. Administrators can control MCP access separately for live mode and sandbox environments in the Dashboard.

Requests and tool parameters go to the Stripe API and are returned to the MCP client and its language model to answer the request. Responses can include payment, customer, and revenue data from the connected Stripe account. The server runs exclusively as a Stripe-hosted service at mcp.stripe.com; there is no self-hostable option.

Because the server allows both read and write API calls (stripe_api_write), a misdirected or manipulated agent could alter real business data, such as triggering payments or overwriting customer records. Stripe itself explicitly recommends enabling human confirmation for tool calls and staying alert to prompt injection when using multiple MCP servers at once, for example via manipulated content from searched documentation. For automated workflows, a restricted API key instead of full account access is recommended, especially for connected accounts.