stripekit mcp

Run stripekit as a Model Context Protocol server, so an AI agent (Claude Desktop, Cursor, Claude Code, …) can reconcile and inspect your Stripe account directly.

npx stripekit mcp

It's a stdio server — the right model for a dev tool: it runs locally, on demand, against your project. There's nothing to host.

Tools

Tool	What it does	Safety
stripekit_plan	Preview the changes push would make.	Read-only
stripekit_check	Verify key, config, drift, and webhook/portal wiring.	Read-only
stripekit_pull	Generate stripe.config.ts from the account.	Read-only
stripekit_push	Reconcile the account to stripe.config.ts.	Dry-run by default. apply: true to actually apply; live: true also required to touch a live account.

Every tool returns structured JSON, and takes an optional cwd (the project directory containing stripe.config.ts).

Configure your agent

Point your MCP client at the server and set cwd to your project (so it finds stripe.config.ts and reads STRIPE_SECRET_KEY from .env.local):

{
  "mcpServers": {
    "stripekit": {
      "command": "npx",
      "args": ["-y", "stripekit@latest", "mcp"],
      "cwd": "/absolute/path/to/your/project"
    }
  }
}

• Claude Desktop: add this to claude_desktop_config.json.
• Cursor: add it to .cursor/mcp.json.
• Claude Code: claude mcp add stripekit -- npx -y stripekit@latest mcp.

If your key isn't in the project's .env.local, pass it through the client's env option instead.

Safety

The write tool mirrors the CLI's guarantees: dry-run unless you ask to apply, a hard live gate for live-mode keys, prices replaced (not mutated), and removals archived (not deleted). Read-only tools can't change anything.