# Leave Management MCP Server

An AI-powered Leave Management System built using the **Model Context Protocol (MCP)**.

This project demonstrates how a Large Language Model (LLM) can discover and invoke MCP tools to perform business operations such as checking leave balances, applying for leave, and retrieving leave history through natural language.

---

## Features

- Built using the FastMCP framework
- Streamlit-based AI client
- Gemini API for intelligent tool selection
- SQLite database for persistent data storage
- Dynamic MCP tool discovery using `list_tools()`
- Natural language interface

---

## Project Structure

```text
.
├── app.py              # Streamlit AI Client
├── main.py             # MCP Server
├── llm.py              # Gemini Integration
├── database.py         # SQLite Helper Functions
├── init_db.py          # Database Initialization
├── pyproject.toml
├── uv.lock
├── .python-version
├── .env.example
├── README.md
└── .gitignore
```

---

## Available MCP Tools

- `get_leave_balance`
- `apply_leave`
- `get_leave_history`

---

## Tech Stack

- Python
- Model Context Protocol (MCP)
- FastMCP
- Streamlit
- Gemini API
- SQLite
- uv

---

## Installation

### 1. Clone the repository

```bash
git clone https://github.com/ratankumarthakur/leave-management-mcp
cd leave-management-mcp
```

### 2. Install dependencies

```bash
uv sync
```

### 3. Configure the environment

Create a `.env` file in the project root.

```text
GEMINI_API_KEY=YOUR_API_KEY
```

### 4. Initialize the database

```bash
python init_db.py
```

### 5. Run the application

```bash
streamlit run app.py
```

---

## Example Queries

Try asking:

- Show leave balance for E001
- Apply leave for E002 on 2026-08-10
- Show leave history for E001
- Apply leave for E001 on 15 September 2026

---

## Architecture

```text
User
   │
   ▼
Streamlit Client
   │
   ▼
Gemini LLM
   │
   ▼
MCP Client
   │
   ▼
MCP Server
   │
   ▼
SQLite Database
```

---

## How It Works

1. The user enters a natural language query.
2. The Gemini model selects the appropriate MCP tool.
3. The Streamlit client invokes the selected MCP tool.
4. The MCP server executes the requested operation.
5. Data is read from or written to the SQLite database.
6. The result is returned to the client and displayed to the user.

---

## Learning Objective

This project was built to understand the fundamentals of the Model Context Protocol (MCP), including:

- Building an MCP server
- Creating an MCP client
- Dynamic tool discovery
- LLM-driven tool invocation
- Database-backed tool execution
- Developing an AI-powered application using Streamlit

--- 

## Screenshot

<img width="1919" height="868" alt="Screenshot 2026-08-03 232719" src="https://github.com/user-attachments/assets/12b8c9b2-6d64-415a-8618-64396af69d61" />