When you connect an AI assistant to a finance app, the scope decides what it can reach, but anything it reads still ends up in your chat. So the privacy check has two halves: grant the narrowest scope the app allows, and ask the assistant only for what you need to see. Start read-only, and disconnect when you are done.
What does a scope actually control?
It controls the app's side of the connection: which actions the assistant is allowed to take and which data it may reach. In SpendixAI you choose a scope when you connect Claude or ChatGPT over MCP, you can start read-only, and you can disconnect whenever you like. The app also keeps records in your own workspace with role-based access, and says it does not sell personal information.
What does a scope not control?
Once the assistant reads a row, the content is part of your conversation. What the assistant's provider keeps, how long, and whether it is used for anything else is governed by that provider's settings and terms, not by the finance app. That is the half of the question many people forget.
| Question | Who controls the answer | Where to look |
|---|---|---|
| Can the assistant change my records? | The app, through the scope | The connection screen |
| What can it read? | The app, through the scope | The scope description |
| How long is the chat kept? | The assistant's provider | The provider's data settings |
| Is my chat used for anything else? | The assistant's provider | The provider's privacy terms |
| Can I take access back? | The app | The disconnect option |
How is this different from the AI categorisation feature?
They are separate. SpendixAI's categorisation works in stages: your own rules first, then your history and similar merchants, and only if AI is on, a model for what is left, using the normalised merchant name and an amount band rather than your name, account numbers, full statement text or balances. You can switch AI off per feature, set a monthly AI budget, or turn it off entirely. An MCP connection is something you choose to start, with an assistant of your choice, and it is not the same pathway.
Some optional features, such as reading an unfamiliar statement layout, do process the uploaded document. Read the in-app description of any feature before turning it on.
What questions should I ask before I approve?
- What exactly is the list of permissions, in words I understand?
- Is there a read-only option, and have I chosen it?
- Which parts of my data will this assistant be able to see?
- What are the chat provider's rules on keeping and training on conversations?
- How do I disconnect, and what does disconnecting remove?
How do I keep what the assistant sees small?
- Ask for totals and trends, not for a list of every payment.
- Avoid asking it to read out account numbers or anything you would not paste into an email.
- Use a separate conversation for each task and delete it afterward.
- Check the scope again after any change to the connector.
- Ask yourself whether you need the assistant at all for the question; the in-app reports may answer it.
As an example, if you ask 'what did I spend on travel in October' and the answer is a single figure such as ₹12,400, the assistant needed to read perhaps a few dozen rows and you needed to see one number. Asking for a total keeps the chat small.
A final test
If you would be uneasy seeing a given detail in a chat that someone else could read over your shoulder, do not ask the assistant to bring it up. Use the app's own screens for that detail instead.
Useful ideas start with a clearer picture of your own spending.
Explore your reports
