AI and MCP

Categorising Uber rides with an AI assistant: what it can see

Before you let an AI assistant tag your Uber rides, here is what it can see, what stays private, how to limit access, and a safer rule-based route.

Editorial illustration for Categorising Uber rides with an AI assistant: what it can see

Letting an assistant tag every Uber ride sounds harmless until you remember that your ride history shows where you live, work and travel. The privacy question is not whether the task is possible, it is how much of your records the assistant must see to do it, and the answer can be very little.

Start from the smallest amount of data

Categorising rides needs a merchant name, a date and an amount. It does not need your name, your account numbers or your balance. A good setup keeps it that way: the assistant lists matching rows, you decide the label, and a rule does the repeat work from then on.

What controls do you have in SpendixAI?

  • Connecting Claude or ChatGPT through MCP is your choice. You pick the scope, you can start read-only, and you can disconnect at any time.
  • There is no bank connection and the app never asks for net-banking logins. It does not read SMS. Sign-in is Google only, so there is no password to leak.
  • Records sit in your own workspace with role-based access, and the product states that it does not sell personal information.
  • Owners can export data, set how long documents are kept, delete transactions and schedule deletion of the whole workspace, with a 7-day window to change their mind.

Which AI is doing what?

These are two different things and worth keeping apart. The first is the built-in categoriser. It works in stages: your rules, then your history and similar merchants, then, only if AI is on, a model for what is left, using a normalised merchant name and an amount band rather than your name, account numbers, full statement text or balances. You can turn this off per feature, set a monthly AI budget, or turn it off entirely, and unmatched rows are then held for review.

The second is an assistant you connect yourself. What it sees follows the scope you grant. Optional features such as reading an unfamiliar statement layout do process the uploaded document, so use those knowingly.

A privacy-first way to do the Uber job

  1. Skip the assistant for the sorting. Filter your transactions for Uber and set the category in bulk by hand.
  2. Add a rule for the merchant name so every later ride is tagged without any assistant or AI involved.
  3. If you do want the assistant, connect it read-only, ask for counts and totals rather than a row-by-row dump, and disconnect afterwards.
  4. Review the connection from time to time and remove any you no longer use.

A small example with made-up numbers

Say 11 Uber rides last month total ₹3,120. You ask the assistant only: how many Uber transactions did I have, and what was the total? It answers 11 and ₹3,120, which matches your own sum, so it read the right rows. You never needed to ask for pickup details or dates, and your statement stays where it is. Then you create the rule yourself.

Questions to ask any tool, including this one

  • Does it need my bank login or OTP? The answer should be no.
  • Can I limit what it reads, and can I cut it off instantly?
  • Where do my documents go, and how long are they kept?
  • Can I export and delete my data myself?
  • Is the AI step optional, and what happens to unmatched rows when it is off?

If a tool cannot answer those in plain words, that is your answer. Keep your records in a place you control, and give an assistant the least access that gets the job done.

SPENDIXAI JOURNAL

Useful ideas start with a clearer picture of your own spending.

Explore your reports