Yes — you stay in control the whole time. When you connect an AI tool (such as Claude or Cursor) to SparkReceipt over MCP, you decide exactly what it can reach and what it can do, and you can cut off its access whenever you want.
Here's what protects you:
- It acts as you, nothing more. The connection is tied to your user and limited to your role's permissions. It can only see and change data you could see and change yourself.
- You approve every connection. Before a tool gets access, a SparkReceipt screen opens in your browser where you choose which accounts to connect and which permission levels to grant.
- Permissions are split into three levels — view data, add and edit data, and irreversible actions — and you grant each one separately. Irreversible actions (which include permanently deleting data) are never pre-selected and carry a clear warning.
- Your password never reaches the AI tool. Sign-in happens on SparkReceipt's own page in your own browser, using a secure standard (OAuth 2.1). The tool only ever receives a limited access token.
- You can disconnect at any time. In Settings → MCP → Connected apps, select Disconnect on any tool and its access ends immediately.
- You can see what it did. The View activity log on the MCP page records every action a connected tool took on your behalf.
One honest caveat: SparkReceipt enforces the permissions you grant, but the AI tool itself decides which actions to take within them. So only connect tools you trust, and grant the minimum you need — especially before you allow irreversible actions.
For the full setup guide and a breakdown of each permission level, see Connect your AI tools to SparkReceipt with MCP.