Expense Tracker MCP Server

Enterprise-grade documentation for a Python-based Model Context Protocol (MCP) expense tracking server.

⚠️ Proof of Concept (PoC)

This is a quick and dirty implementation based on frameworks and libraries available as of February 2026.

Behavior, APIs, and integration patterns may evolve in future versions of FastMCP, OpenAI Codex, and related tooling. This is just for demonstrative purposes

Table of Contents

1. Introduction
2. Architecture Overview
3. Environment Setup (uv-based)
4. Installed Packages
5. Project Structure
6. MCP Server Overview
7. Core Functional Components
8. Excel Storage Layer
9. Running the Server
10. Testing with MCP Inspector
11. Integrating with Codex in VS Code
12. Operational Considerations
13. Future Improvements

1. Introduction

This project implements a local Model Context Protocol (MCP) server using Python and FastMCP.

📢 Educational Proof of Concept

This repository contains intentionally simple and demonstrative code designed to get started with MCP servers and understand how they work. The implementation prioritizes clarity and approachability over production-grade architecture, advanced patterns, or highly optimized design.

The server allows natural-language expense logging such as:

"I spent 20 dollars on a Batman figure yesterday"

The server:

• Parses the amount
• Detects currency
• Extracts relative or explicit dates
• Stores the result in an Excel file
• Exposes tools via MCP for AI agents (e.g., OpenAI Codex in VS Code)

This project demonstrates:

• Local stdio-based MCP server design
• Natural language parsing
• Structured data persistence
• Tool registration via FastMCP
• Integration with OpenAI Codex UI

Example result in Codex on VS Code

2. Architecture Overview

The solution consists of:

• FastMCP stdio server
• Natural language parser
• Excel persistence layer (openpyxl)
• Tool registration via decorators
• Local stdio transport for MCP communication

Transport Type: - STDIO (standard input/output)

Data Storage: - Excel file (expenses.xlsx)

Execution Model: - Event-driven tool invocation

3. Environment Setup (uv-based)

The environment was configured using uv for fast dependency management.

Initialize project

uv init

Add required packages

uv add fastmcp openpyxl dateparser pypandoc

This creates:

• Virtual environment
• Dependency resolution
• Lockfile
• Reproducible environment

4. Installed Packages

Package Purpose

fastmcp MCP server implementation openpyxl Excel read/write operations dateparser Natural language date parsing pypandoc Documentation generation re Regex amount parsing pathlib File handling datetime Timestamp management

5. Project Structure

project-root/
│
├── server.py
├── expenses.xlsx
├── README.md
└── .venv/

6. MCP Server Overview

The server is initialized as:

mcp = FastMCP("expense-tracker")

Tools are registered using:

@mcp.tool

The server starts via:

if __name__ == "__main__":
    mcp.run()

The server communicates using STDIO and must not print to stdout.

7. Core Functional Components

7.1 _parse_amount(raw)

Extracts numeric amount using regex. Handles: - 12.000,16 - 12,000.16 - 12000

Returns:

float

7.2 _parse_currency(raw)

Detects: - $ → USD - € → EUR - £ → GBP - keyword matches (dollars, euro, etc.)

Returns ISO currency code.

7.3 _parse_date_iso(raw)

Uses:

dateparser.search.search_dates()

Configuration: - PREFER_DATES_FROM = "past" - RELATIVE_BASE = datetime.now() - RETURN_AS_TIMEZONE_AWARE = False

Returns:

YYYY-MM-DD

Fallback: - If no date detected → today

7.4 _parse_expense(text)

Combines: - amount - currency - date - description

Returns structured dictionary:

{
    "amount": float,
    "currency": str,
    "date_iso": str,
    "description": str
}

8. Excel Storage Layer

Excel file created if missing:

expenses.xlsx

Header structure:

["Date", "Description", "Amount", "Currency", "Raw Text", "Logged At"]

Append logic ensures: - Workbook exists - Correct sheet name - ISO timestamp logging

Read logic: - Dynamically maps header row - Avoids tuple index errors - Skips blank rows

9. Running the Server

Direct execution

.\.venv\Scripts\python.exe server.py

The process should remain running (stdio server).

10. Testing with MCP Inspector

Launch:

npx @modelcontextprotocol/inspector python server.py

Steps: 1. Open browser UI 2. View tools 3. Call log_expense 4. Call list_expenses

11. Integrating with Codex in VS Code

Open Codex MCP UI

"Connect to a custom MCP"

Select: - STDIO

Configuration

Command to launch:

C:\Users\andre\Documents\Python\MCP\Python\Stdio Server\.venv\Scripts\python.exe

Arguments:

C:\Users\andre\Documents\Python\MCP\Python\Stdio Server\server.py

No environment variables required.

After saving: - Enable the MCP tool - Use in Codex chat:

Example:

Log this expense: I spent 50 dollars on groceries yesterday

12. Operational Considerations

• Do not print to stdout
• Use stderr for debugging
• Always use absolute python path
• Ensure virtual environment consistency
• Keep Excel closed during writes
• Consider file locking for production use

13. Future Improvements

• Category auto-detection
• Deduplication logic
• CSV export
• SQLite backend
• Multi-user storage
• Authentication layer
• Cloud deployment (HTTP MCP)
• Structured validation with Pydantic