An expense tracker API lets another program add transactions and read them back, using a secret key instead of your login. You never need to write code to understand it: think of the key as a staff pass that only opens two doors, one to put entries in and one to look at them. This tutorial explains the idea in plain words so you can tell what a developer or vendor is promising.
What is an API, in kitchen terms?
A restaurant has a menu, a counter and a kitchen. You do not walk into the kitchen; you place an order at the counter and get a plate. An API is the counter. The expense app is the kitchen, and your request is the order. Requests are short structured messages, and the replies come back in the same structure.
What does creating a transaction look like?
A create request carries the same facts you would type into a form. Written the way a developer sees it, with made-up values, it is a small list of fields.
| Field | Example value | Plain meaning |
|---|---|---|
| date | 2026-11-05 | When the payment happened |
| amount | 1450.00 | How much, as a number without the rupee sign |
| currency | INR | Which currency the amount is in |
| merchant | Sharma Stationers | Who was paid |
| category | Office supplies | Where it belongs in your books |
| note | Chart paper for the workshop | A reminder for later |
If everything is right, the app replies with the saved record and an identifier. If something is missing or malformed, it replies with an error that names the field. A beginner's rule: an error message is a free lesson, so read it before retrying.
What does reading transactions look like?
A read request asks for records, usually with a filter such as a date range or a category. The reply is a list of the same fields. A finance user would use this to pull the month's rows into a sheet or a report without downloading a file each time.
Where does the key come in?
The key is a long string that proves the request is yours. Treat it exactly like an ATM PIN, with one difference: it can be generated again. If it leaks, replace it and the old one stops working.
- Keep it in the tool that needs it, not in an email, a chat message or a shared sheet
- Give each tool its own key so you can switch one off without breaking the rest
- Prefer a key that can only read if the job is only reading
- Never share your net-banking login or an OTP to make an integration work; a proper API never asks for either
Try it on paper first: a hypothetical month
Say a tuition teacher takes ₹6,000 from each of five students by UPI and spends ₹2,400 on books and ₹900 on printing. Her helper script would create five income records and two expense records. Reading them back for the month should give ₹30,000 in, ₹3,300 out, and ₹26,700 net (30,000 minus 3,300). If the total from the API differs, either a record failed to save or one was entered twice.
What should you check before trusting an API?
- Is the format of dates and amounts written down, including how paise are handled?
- Can a request be repeated safely, or will a retry create a duplicate?
- Are there limits on how many records you can send in a minute or a month?
- Can you delete or fix a record afterwards, and is there a recycle bin?
- Can a key be limited to reading only, and can you revoke it?
SpendixAI's pages describe statement import and manual entry as the ways records arrive, and a connection for Claude or ChatGPT through MCP. Check its own documentation for anything beyond that, rather than assuming a feature exists.
When to ask someone technical
If a vendor demo needs your bank login, stop. If a script will run unattended on your accounts, ask a developer to log what it sends and to add a way to switch it off. Beyond that, you now have enough vocabulary to ask sensible questions.
Useful ideas start with a clearer picture of your own spending.
Explore your reports
