Connect Cursor. Then test your app.
Start with a working connection. Your agent can then inspect the app, define its checks and show what actually happened.
1. Find your command and get a real key.
On Mac or Linux, run command -v npx in Terminal. On Windows, use where.exe npx. Use the full path for your installed Node.js command. Create your account key and keep it private.
The example below was tested with Homebrew on Apple Silicon Mac and MCP 0.7.24. Replace the command if your path differs. Replace fsk_your_key_here with your real raw key, without a Bearer prefix. The placeholder cannot authenticate.
2. Save the config and restart Cursor.
Use one FetchSandbox entry in ~/.cursor/mcp.json. Check for a conflicting entry in your project’s .cursor/mcp.json.
{
"mcpServers": {
"fetchsandbox": {
"command": "/opt/homebrew/bin/npx",
"args": ["-y", "[email protected]"],
"env": { "FETCHSANDBOX_API_KEY": "fsk_your_key_here" }
}
}
}Fully quit and reopen Cursor. Reloading the window alone may keep the old MCP process. Open its MCP settings, enable FetchSandbox if needed and confirm the tools are listed. A connected server is not yet a verified app.
3. Ask it to validate the running app.
Use your own app, or start with an existing playground app. For a Stripe and Resend app, ask:
Use FetchSandbox to validate this app’s Stripe and Resend integration. Test it, fix any issues, and show me the receipt.
The agent should continue unfinished applicable checks or show the exact blocker. The receipt records what passed, failed or remains unmeasured. It does not promise every production failure is covered. Your account key connects MCP; temporary provider and webhook credentials are separate and belong in the app’s private test setup.
A startup error points to the command or Node installation. A 401 during verification points to authentication. Check the error in Cursor’s MCP output before changing the setup. See the engine session reference or return to the docs overview.