mikhilv/mcp_server_new

MCP Server for Natural Language SQL Queries

This project implements a Model Context Protocol (MCP) server that accepts natural language queries, converts them to SQL using machine learning, executes them on a SnowSQL database, and returns the results in JSON format.

Setup

1. Clone the repository
2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

4. Configure your SnowSQL connection:
   • Copy .env.example to .env
   • Update the values in .env with your SnowSQL credentials

Running the Server

To run the server:

uvicorn src.server:app --reload

The server will start on http://localhost:8000

API Usage

POST /query

Send natural language queries to get SQL results.

Request body:

{
    "query": "Show me all sales from last month"
}

Response:

{
    "sql_query": "SELECT * FROM sales WHERE date >= DATEADD(month, -1, CURRENT_DATE())",
    "results": [
        {
            "sale_id": 1,
            "amount": 100.50,
            "date": "2023-09-15"
        }
        // ... more results
    ]
}

Security Notes

• The server only allows SELECT queries for security reasons
• Make sure to properly configure your SnowSQL user permissions
• Keep your .env file secure and never commit it to version control

Error Handling

The server provides detailed error messages for:

• Invalid natural language queries
• SQL generation failures
• Database connection issues
• Query execution errors