Personal Finance MCP Server

"Wait, where did my money go this month?"

We've all been there. You check your bank balance at the end of the month and it's somehow $400 less than you expected. You open your transactions, scroll for five minutes, and still can't figure out what happened.

This project fixes that — by giving Claude the ability to actually analyze your finances. Not just read numbers, but understand them, flag the problems, and tell you exactly what to do differently.

The idea

Most finance apps are dashboards you have to go to. This is different — it's an MCP server that plugs directly into Claude Desktop, so your AI assistant gains finance analysis as a native ability. You just talk to it.

"Analyze my spending from last month and tell me where I'm bleeding money."

Claude pulls your transactions, categorizes everything, compares it against the 50/30/20 rule, surfaces the problem areas, and gives you a plain-English report with actionable tips — all in one response, no spreadsheets, no dashboards.

Related MCP server: Finance Tracker MCP Server

What's under the hood

Eight tools that chain together:

Real bank data via Plaid

The CSV flow works with any bank export. But if you want live data, the Plaid integration connects to actual bank accounts (Chase, BofA, Wells Fargo, 12,000+ others) and pulls real transactions in real time.

For testing, Plaid's sandbox gives you a fake-but-realistic bank with pre-loaded transactions — no real account needed.

Get it running

1. Clone and set up:

git clone https://github.com/jahnavi-reddy03/personal-finance-mcp.git
cd personal-finance-mcp
python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # Mac/Linux
pip install -r requirements.txt

2. Add your API keys:

cp .env.example .env
# then edit .env and fill in your OpenAI + Plaid keys

3. Wire it into Claude Desktop — add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "personal-finance": {
      "command": "C:\\path\\to\\venv\\Scripts\\python.exe",
      "args": ["C:\\path\\to\\personal-finance-mcp\\server.py"]
    }
  }
}

Restart Claude Desktop. The eight tools load automatically — you'll see them in Settings → Developer → Local MCP Servers.

Try it yourself

There's a 30-transaction sample CSV in data/sample/transactions.csv so you can test immediately without connecting a bank. Just paste this into Claude Desktop:

Analyze my spending from C:\path\to\personal-finance-mcp\data\sample\transactions.csv,
flag anything over budget, and give me a full report with savings tips.

For live Plaid data, run the helper script first to get a sandbox access token:

python get_sandbox_token.py

Tech stack

• Python + FastMCP — the MCP server itself and all tool definitions

• Pandas — transaction parsing, categorization, and aggregation

• OpenAI GPT-3.5 — generates tips and report narrative (falls back to static tips if no API key)

• Plaid API — live bank transaction fetching across 12,000+ institutions

• python-dotenv — keeps API keys out of the codebase

Project structure

personal-finance-mcp/
├── server.py              ← entry point, all 8 tools registered here
├── tools/
│   ├── analyze.py         ← CSV parsing + keyword categorization
│   ├── overspending.py    ← 50/30/20 benchmark comparisons
│   ├── tips.py            ← GPT-3.5 tips + static fallback
│   ├── report.py          ← plain-English report generator
│   ├── plaid_fetch.py     ← live bank data via Plaid
│   ├── predict_balance.py ← month-by-month balance projection
│   ├── spending_patterns.py ← weekly pattern analysis + large purchases
│   └── savings_plan.py    ← GPT-3.5 3-month savings plan
├── data/sample/
│   └── transactions.csv   ← 30 realistic test transactions
├── docs/
│   └── demo.gif           ← screen recording of live tool usage
├── get_sandbox_token.py   ← one-time script to get Plaid sandbox token
├── .env.example           ← copy this to .env, never commit .env
├── requirements.txt
└── LICENSE

Built by

Jahnavi Reddy

License

MIT — use it, fork it, build on it.