AI and MCP

Personal Finance MCP Server for Claude: A Privacy Check Before You Connect

A privacy-first checklist for connecting a personal finance MCP server to Claude: what is shared, scope limits, AI controls, retention, deletion and disconnecting.

Editorial illustration for Personal Finance MCP Server for Claude: A Privacy Check Before You Connect

Before you connect any personal finance MCP server to Claude, you should know exactly which of your records the assistant can request, whether it can change anything, and how to remove the connection. If the app cannot answer those three questions clearly, wait. This is a privacy checklist, written for someone who would rather ask first.

What actually travels to the assistant?

When you ask Claude a question, the app answers with the rows or totals needed for that question, and that answer becomes part of your conversation. So the privacy question is really two: what the app will hand over, and what you are comfortable having in a chat history. Ask for the narrowest answer that does the job.

The checklist

QuestionWhat a good answer sounds like
Can I limit access to reading?Yes, and I can widen it later
Can I see who is connected?A list with a disconnect option
Is my data sold?A plain statement that personal information is not sold
Does the app ever need my bank login?No; records come from statements I upload
What if I delete something by mistake?A recycle bin with a time window
Can I leave completely?Export, then scheduled deletion with time to change my mind

What does SpendixAI say about itself?

SpendixAI does not connect to your bank, never asks for net-banking logins, does not read SMS, and uses Google sign-in only. Your records live in your own workspace with role-based access, it states that it does not sell personal information, and owners can export data, choose document retention, delete transactions and schedule deletion of the whole workspace with a 7-day window to change their mind. Claude or ChatGPT can be connected via MCP: you choose a scope, can start read-only and can disconnect any time.

A worked example of limiting exposure

Say you want to know whether eating out has crept up. A broad request such as send me everything for the year might return 1,400 rows. A narrow one, total of the eating-out category by month for six months, returns six numbers. Suppose they read ₹2,800, ₹3,100, ₹2,900, ₹3,600, ₹4,200 and ₹4,500. The last three average ₹4,100 (12,300 divided by 3) against ₹2,933 for the first three (8,800 divided by 3), a rise of about ₹1,167 a month. Six numbers answered the question; fourteen hundred rows would not have.

  1. Connect read-only
  2. Ask narrow questions by category and month
  3. Avoid putting names, notes with health details or ID numbers in prompts
  4. Review the connection list after each session
  5. Disconnect if you will not use it for a while

How are the AI features controlled?

In the app, categorisation runs in stages: your rules first, then your history and similar merchants, and only if AI is on, a model for what remains, using the normalised merchant name and an amount band rather than your name, account numbers, full statement text or balances. AI can be switched off per feature, capped with a monthly budget or turned off entirely, in which case unmatched items are held for review. Optional features such as reading an unfamiliar statement layout do process the uploaded document.

What if you change your mind later?

Plan the exit before you connect. Know where the list of connections lives and practise disconnecting once with a test question afterwards. Decide how long you are happy for past conversations to stay in the assistant, and delete the ones that contained rows you now regret sharing. A connection you can leave in a minute is far easier to accept than one you are unsure how to end.

Closing routine

  • Write down which assistant is connected and why
  • Revisit the list monthly
  • Export a copy of your data before any big change
  • Delete test entries you created while experimenting
SPENDIXAI JOURNAL

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

Explore your reports