Personal Finance

💰 Personal Finance MCP

Deterministic personal-finance toolkit exposed over the Model Context Protocol — 77 calculators, a meta-advisor, and live market data, with a polished web UI. Grounded in established financial mathematics.

[Python 3.10+] [MCP] [Tests] [License: MIT] [Live on Hugging Face]

Live demo: https://sarveshtalele-personal-finance-mcp.hf.space Connector URL: https://sarveshtalele-personal-finance-mcp.hf.space/mcp

Demo

[Watch the demo]

▶️ Watch the 2-minute demo — plain-language question → chained tools → a prioritized plan.

Public demo note: the hosted Space is a shared, best-effort instance (rate-limited, may cold-start after idle). For heavy or private use, run it locally or self-host (see docs/HOW_IT_WORKS.md).

Overview

Most finance "assistants" guess at numbers. This one doesn't. It ships 77 deterministic calculators — same inputs, same answer, every time — and lets an LLM route a plain-language question to the right tools. Describe your situation ("I'm 30, earn ₹1L/month, want to retire at 60") and the create_financial_plan orchestrator chains the relevant calculators into a single prioritised plan.

It runs three ways from one codebase:

• As an MCP server — connect it to Claude Desktop, Claude Code, Cursor, or any MCP client.
• As a website — a Next.js UI with a live calculator, a market dashboard, and a tool catalog.
• As a hosted connector — deployed to a Hugging Face Docker Space; one URL does all three.

Highlights

• 🔢 Deterministic — pure math, no model inference for the numbers.
• 🤖 Story → tools — the model maps intent to tools; users never name them.
• 🇮🇳 Theory-grounded — TVM, debt, PPF/SSY/NSC/EPF, bonds, derivatives, MPT, and more.
• 🛰️ Live market data — mutual-fund NAVs (AMFI), FX (ECB), equity quotes (Yahoo) — no API keys.
• 🔒 Hardened — stateless, rate-limited, input-bounded APIs with security headers/CSP.

Tool catalog — 77 tools, 13 categories

Browse them all (with live descriptions) at /tools.

Quick start

Use the hosted connector (no install)

Claude Desktop — Settings → Connectors → Add custom connector → paste:

https://sarveshtalele-personal-finance-mcp.hf.space/mcp

Claude Code

claude mcp add --transport http personal-finance https://sarveshtalele-personal-finance-mcp.hf.space/mcp

Cursor / VS Code — add to mcp.json:

{  "mcpServers": {    "personal-finance": {      "url": "https://sarveshtalele-personal-finance-mcp.hf.space/mcp",      "transport": "http"    }  }}

Install from PyPI (stdio server)

pip install personal-finance-mcp     # or: uvx personal-finance-mcp

Then point Claude Desktop at it:

{  "mcpServers": {    "personal-finance": { "command": "uvx", "args": ["personal-finance-mcp"] }  }}

Run locally from source

git clone https://github.com/sarveshtalele/personal-finance-mcp.gitcd personal-finance-mcppip install -e .
# Option A — classic stdio MCP server (offline, no web)python -m src
# Option B — unified server: website + /mcp connector + /api  (http://localhost:7860)cd web && npm install && npm run build && cd ..python -m src.web

For stdio, point Claude Desktop at the local process:

{  "mcpServers": {    "personal-finance": { "command": "python", "args": ["-m", "src"] }  }}

The website

python -m src.web serves everything on one port:

Architecture

src/├── server.py            # FastMCP server — registers all tool modules├── __main__.py          # `python -m src`  (stdio transport)├── tools/               # pure math fns + per-module register(mcp)│   ├── tvm.py  debt.py  planning.py  bonds.py  stocks.py  mutual_funds.py│   ├── portfolio.py  derivatives.py  india_savings.py  cashflow.py│   ├── risk_profile.py  advisor.py   # advisor = story → plan orchestrator│   └── marketdata.py    # live AMFI / Frankfurter / Yahoo (keyless)├── models/              # Pydantic schemas + enums├── utils/               # output formatters└── web/                 # unified Starlette server (MCP + /api + static site)    ├── server.py        # routes, security middleware, calculator registry    └── __main__.py      # `python -m src.web`  (uvicorn, port 7860)
web/                     # Next.js front-end (static export → web/out)└── app/                 # home, tools, calculator, dashboard, connect

Each tool file keeps deterministic pure functions separate from the thin @mcp.tool wrappers, so the same functions power the MCP server, the web calculators, and the tests.

Security

• Stateless — no database, no sessions; every call is independent and reproducible.
• Hardened API — per-IP rate limiting, request-body cap, and input validation that bounds loop-driving parameters (years/months/age) to prevent denial-of-service.
• Security headers — CSP (with frame-ancestors for the Hugging Face embed), X-Content-Type-Options, Referrer-Policy, Permissions-Policy; CORS limited to GET/POST without credentials. The header middleware is implemented at the ASGI layer so it never buffers the streaming /mcp (SSE) responses.
• No secrets in the app — live-data sources are public and keyless.

See SECURITY.md to report a vulnerability.

Development

pip install -e ".[dev]"pytest -q                       # 119 testsruff check .                    # lintpython -m src.web               # run the full stack locally

Deployment

Deployed as a Hugging Face Docker Space, auto-synced from GitHub on every push to main (see .github/workflows/hf-sync.yml). Full instructions — local, Docker, and Hugging Face — are in docs/deployment.md.

Documentation

• docs/HOW_IT_WORKS.md — concepts (MCP, transports, semantic routing), full architecture with diagrams, end-to-end request flows, the security model, hosting it yourself / on a portfolio site, and a production-grade roadmap.
• docs/deployment.md — local, Docker, and Hugging Face deployment.
• docs/Architecture.md · docs/testing.md · docs/setup.md

Contributing

Contributions are welcome — see CONTRIBUTING.md and the Code of Conduct. Good first issues: add a calculator (a pure function + a register wrapper + a test), improve descriptions for better tool routing, or extend the web UI.

Disclaimer

Educational tool for illustrating standard financial formulas. Not investment advice. Figures are illustrative; verify before making financial decisions.

License

MIT — free to use, modify, and distribute.