A multi-user AI-powered expense tracking server built with FastMCP + FastAPI + PostgreSQL. Users can manage finances using plain English with AI.

Works with:

• 🖥 Claude Desktop
• 📱 Claude.ai Free (via N8N)
• 🌐 Any HTTP client

---

✨ Features

• 🔐 JWT Authentication – secure multi-user system
• 👥 Multi-user support – each user has isolated data
• 🧰 29 MCP tools for expenses, income, budgets and analytics
• 🌐 Mobile friendly registration page
• ☁ Cloud deployable (Railway + Neon)
• 🤖 AI integration with Claude
• 🔄 Claude.ai free support using N8N bridge

---

📁 Project Structure

expense_tracker_v3/

├── CORE
│   ├── main.py
│   ├── run.py
│   ├── app.py
│   ├── config.py
│   ├── context.py
│   ├── db.py
│   ├── init_db.py
│   ├── logger.py
│   ├── utils.py
│   └── create_user.py
│
├── DATABASE
│   ├── schema.sql
│   └── categories.json
│
├── API
│   └── api/
│       ├── __init__.py
│       ├── auth.py
│       ├── middleware.py
│       └── server.py
│
├── MCP TOOLS
│   └── tools/
│       ├── expense_tools.py
│       ├── income_tools.py
│       ├── budget_tools.py
│       ├── category_tools.py
│       ├── summary_tools.py
│       └── utility_tools.py
│
├── FRONTEND
│   └── static/
│       └── register.html
│
├── DEPLOYMENT
│   ├── Procfile
│   ├── pyproject.toml
│   └── requirements.txt
│
└── README.md

⚠ Never commit .env to GitHub.

---

🧰 MCP Tools (29)

Expense Tools

• add_expense
• update_expense
• delete_expense
• list_expenses
• get_expense_by_id

Income Tools

• add_income
• list_income
• delete_income
• monthly_income

Budget Tools

• set_budget
• get_budget
• check_budget_status
• delete_budget

Category Tools

• get_categories
• add_category
• update_category
• delete_category

Summary Tools

• summarize_expenses
• daily_summary
• weekly_summary
• monthly_summary
• yearly_summary
• category_breakdown
• top_spending
• compare_months
• get_balance

Utility Tools

• get_last_expenses
• search_expenses
• export_expenses_csv

---

🗄 Database Schema

users
 id
 username
 password
 created_at

categories
 id
 name

expenses
 id
 user_id
 date
 amount
 category_id
 note

income
 id
 user_id
 date
 amount
 source

budgets
 id
 user_id
 category_id
 monthly_limit

All queries automatically filter:

WHERE user_id = current_user

Each user has completely isolated data.

---

🚀 Local Setup

1️⃣ Clone Repository

git clone https://github.com/parnajaswanth227/Expense_Tracker_With_Claude.git
cd Expense_Tracker_With_Claude

---

2️⃣ Create Python Environment

uv init
uv venv --python 3.12
.venv\Scripts\activate

Windows users:

$env:UV_LINK_MODE="copy"

---

3️⃣ Install Dependencies

uv add fastmcp fastapi uvicorn psycopg[binary] python-dotenv python-jose bcrypt
uv pip install -r requirements.txt

---

4️⃣ Create .env

Create a .env file in the root folder.

DATABASE_URL=postgresql://user:pass@host/dbname?sslmode=require
SECRET_KEY=your_secret_key_here
ALLOW_REGISTRATION=true
ACCESS_TOKEN_EXPIRE_MINUTES=525600

Generate a secret key:

python -c "import secrets; print(secrets.token_hex(32))"

---

5️⃣ Start Server

uv run python run.py

Server runs at:

http://localhost:8000

---

🔑 API Endpoints

Method	Endpoint	Description
GET	/health	Server status
GET	/register	Signup page
POST	/auth/register	Create user
POST	/auth/token	Login
POST	/mcp	MCP endpoint

---

☁ Cloud Deployment

Step 1: Create PostgreSQL Database

Go to:

https://neon.tech

Create database and copy connection string.

---

Step 2: Deploy on Railway

Go to:

https://railway.app

Create project → Deploy from GitHub.

Railway will automatically use:

web: uvicorn api.server:app --host 0.0.0.0 --port $PORT

Add environment variables:

Variable	Value
DATABASE_URL	Neon database URL
SECRET_KEY	generated secret
ALLOW_REGISTRATION	true
ACCESS_TOKEN_EXPIRE_MINUTES	525600

---

Step 3: Verify Deployment

https://your-app.railway.app/health

Expected response:

{"status":"ok"}

---

🖥 Using with Claude Desktop

Install MCP server:

uv run fastmcp install claude-desktop main.py

Manual configuration:

%APPDATA%\Claude\claude_desktop_config.json

{
 "mcpServers": {
   "ExpenseTracker": {
     "command": "npx",
     "args": [
       "-y",
       "mcp-remote@latest",
       "https://your-app.railway.app/mcp",
       "--header",
       "Authorization: Bearer YOUR_TOKEN"
     ]
   }
 }
}

---

📱 Using Claude.ai Free (Mobile + Web)

Claude free plan does not support MCP connectors directly. We use an N8N workflow bridge.

N8N Architecture

Claude AI
   ↓
N8N Webhook
   ↓
Expense Tracker MCP
   ↓
PostgreSQL Database

---

Setup N8N

Step 1: Create Account

https://n8n.io

Create account.

---

Step 2: Create Workflow

Click:

New Workflow

---

Step 3: Import Workflow JSON

Click:

⋮ → Import from JSON

Paste provided workflow.

---

Step 4: Add Token

Replace:

YOUR_JWT_TOKEN_HERE

with your token.

---

Step 5: Activate Workflow

Click Activate.

You will receive webhook URL:

https://yourname.app.n8n.cloud/webhook/expense-tracker

---

Connect N8N to Claude.ai

1. Open Claude.ai
2. Go to:

Settings → Connectors

3. Select N8N
4. Paste your webhook URL.

Claude can now control your expense tracker.

---

Example Commands

Users can ask Claude:

Add expense ₹500 for lunch today

Show my expenses this month

Add ₹50,000 salary income

Show my balance

Check my budget

Claude will automatically call MCP tools.

---

👥 Multi-User Flow

User opens /register
      ↓
Creates username & password
      ↓
Receives JWT token
      ↓
Token used in
 - Claude Desktop
 - N8N workflow
 - API calls

Each user has separate database data.

---

🔧 Admin Create Users

python create_user.py --username alice --password pass123

Disable public signup:

ALLOW_REGISTRATION=false

---

🐛 Troubleshooting

Error	Solution
401 Unauthorized	Token expired
Database error	Check DATABASE_URL
MCP session error	Restart server
N8N webhook error	Check workflow JSON

---

🔐 Security

Never commit:

.env
JWT tokens
database credentials

Rotate tokens if exposed.

---

📦 Tech Stack

Layer	Technology
MCP Framework	FastMCP
API	FastAPI
Database	PostgreSQL
Cloud DB	Neon
Deployment	Railway
Automation	N8N
AI Client	Claude

---

⭐ Project Goal

Create a fully AI-powered personal finance assistant where users manage expenses using natural language with Claude AI.

---

If you'd like, I can also help add:

• 🧭 Architecture diagrams
• 📸 Screenshots
• 🔄 Workflow diagrams
• 🏷 GitHub badges
• 🎬 Animated demo

to make your GitHub repository look like a professional open-source project.

📸 Screenshots

screenshorts\Screenshot 2026-03-14 113116.png

💻 Web Dashboard – Expense Summary

The AI assistant analyzes your expenses and presents a clear financial overview including income, spending distribution, and balance.

Features shown:

• Total income, expenses, and balance
• Category-wise spending breakdown
• Interactive spending chart
• AI-generated financial insights

Example insight:

You've spent ₹2,000 across 3 transactions. Food is the dominant category at 85% (₹1,700). With ₹60,000 income, you're saving 96.7% — a healthy balance of ₹58,000.

---

📱 Claude AI Mobile – Natural Language Expense Entry

C:\Users\parna\Videos\expense_tracker_v3\screenshorts\WhatsApp Image 2026-03-14 at 11.33.34 AM.jpeg Users can add expenses using natural language with AI.

Example command:

Add an expense 210 rs for petrol for my bike

Claude automatically:

1. Detects the transport category
2. Calls the add_expense MCP tool
3. Stores the expense in the database
4. Returns confirmation with details

Returned response includes:

• Expense ID
• Amount
• Category
• Note
• Date

This demonstrates how users can manage finances conversationally with AI.

🌐 Live Demo

The project is deployed on Railway Cloud.

📝 User Registration Page

Create a new account:

https://web-production-d3df0.up.railway.app/register

This page allows users to:

• Create a username
• Set a password
• Generate a JWT token for authentication

---

🔌 MCP Server Endpoint

Main MCP endpoint used by Claude and N8N:

https://web-production-d3df0.up.railway.app/mcp

This endpoint handles:

• MCP tool execution
• AI requests from Claude
• Automation requests from N8N workflows

---

🧪 Health Check (Optional)

Check if the server is running:

https://web-production-d3df0.up.railway.app/health

Expected response:

{"status":"ok"}