AI and MCP

Categorise all Uber rides with an AI assistant: a beginner tutorial

A plain-language tutorial for non-technical users: get an AI assistant to tag Uber rides as travel and Uber Eats as food, with a small worked example.

Editorial illustration for Categorise all Uber rides with an AI assistant: a beginner tutorial

If you spend your day in Excel and bank statements rather than code, this is for you. You can ask an AI assistant to find every Uber charge in your expense records and file the rides under Travel, and the food orders under Food. The trick is to give the instruction in small, checkable pieces instead of one big command.

What are we actually setting up?

There are two parts. First, your expense tracker holds your transactions, imported from bank and card statements or typed in. Second, an assistant such as Claude or ChatGPT is connected to that workspace. The connection method is called MCP, which is just a standard way for an assistant to talk to an app. You decide what the assistant is allowed to touch, you can start with look-only access, and you can disconnect it any time.

Step by step

  1. Open your tracker and check that the month's statements are imported. If a card statement is missing, the assistant cannot see those charges.
  2. Pick the account first when you import. Each transaction keeps its account, which helps later when you wonder which card a ride was charged to.
  3. Connect the assistant and choose read-only for now.
  4. Type a simple request such as: show me every transaction with Uber in the description, with date, amount and current category.
  5. Read the answer. Count the rows, then add the amounts on a calculator.
  6. Only then ask for the change, or make it yourself in the app if you chose to keep the assistant read-only.

A small example to practise on

Imagine a statement with five Uber-looking lines: ₹180, ₹240, ₹310, ₹95 and ₹460. The ₹460 one is a food delivery, even though the name on the statement starts with the same word. You tell the assistant: rides go to Travel, anything that is a food order goes to Food.

LineAmountWhere it should go
Ride₹180Travel
Ride₹240Travel
Ride₹310Travel
Ride₹95Travel
Food order₹460Food

Travel should come to 180 + 240 + 310 + 95 = ₹825. Food should be ₹460. The grand total is ₹1,285. If your tracker shows different totals after the change, something went to the wrong place.

What if we do not understand how the assistant decided?

Ask it. A good request is: explain which rows you changed and why. If it cannot tell you, undo the change by correcting the rows yourself. Deleted transactions in SpendixAI sit in a recycle bin for 30 days, so a slip with a delete is recoverable, but it is still far simpler to review before changing.

Do I need to repeat this every month?

No. Once the first month is tidy, create a rule so the tracker files Uber rides as Travel on its own. SpendixAI applies your own rules first, before history or any AI, so the result stays predictable. Corrections you make also become history, and you can add a merchant alias if one shop appears under several names.

Questions beginners usually have

  • Will this change my bank balance? No. It only changes how rows are labelled inside the tracker.
  • Can I do it without an assistant? Yes. Filter the transactions by the word Uber and change the category in bulk by hand, or add a rule.
  • What if the assistant gets something wrong? Correct it yourself and carry on. Treat its answers as a first draft.
  • Is it fine to ask in plain English? Yes, but specific requests work better than vague ones, so name the month, the category and the account.

Once you have done this one merchant, repeat it for the next regular spender in your statements, such as a metro card top-up or a streaming service, one at a time.

SPENDIXAI JOURNAL

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

Explore your reports