SQL MCP Server

The SQL MCP Server is a unified agentic interface that allows AI models (like Claude) to inspect, query, and interact with multiple types of databases through a single, standardized set of tools.

Instead of installing and configuring separate servers for MySQL, SQL Server, Oracle, etc., this single package handles them all by dynamically loading the correct "adapter" based on your configuration.

🚀 Features

• Standardized Toolset: The AI always sees the same tools (execute_query, describe_table, get_db_version), no matter if it's talking to Oracle or MySQL. This reduces hallucinations and improves query accuracy.

• Resources: Automatically exposes database tables as MCP Resources for easy inspection.

• Smart Adapters:

  • ✅ MySQL: Uses mysql-connector-python with connection pooling.

  • ✅ MSSQL: Uses pymssql and auto-detects schema names (e.g., dbo.Users).

  • ✅ Sybase / SAP IQ: Uses pyodbc and intelligently handles dba. prefixes and server names.

  • ✅ Oracle: Uses oracledb (Thin mode) for easy connectivity without complex driver installs.

• Unicode Ready: All connections default to UTF-8 to handle special characters seamlessly.

Related MCP server: Universal SQL MCP Server

🛠️ Installation

1. Clone the repository:

   git clone https://github.com/BryR0/sql-mcp.git
   cd sql-mcp

2. Create a virtual environment:

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

3. Install dependencies:

   pip install -e .

⚙️ Configuration

You define which database to use via the DB_TYPE environment variable.

Supported Environment Variables

Claude Desktop Config Example

Add this to your claude_desktop_config.json. Note that DB_PORT is optional and will default to the standard port for the chosen DB type if omitted.

{
  "mcpServers": {
    "sql_mcp_mysql": {
      "command": "C:\\path\\to\\venv\\python.exe",
      "args": ["-m", "sql_mcp"],
      "env": {
        "DB_TYPE": "mysql",
        "DB_HOST": "localhost",
        "DB_USER": "root",
        "DB_PASSWORD": "password",
        "DB_DATABASE": "my_db"
      }
    },
    "sql_mcp_mssql": {
      "command": "C:\\path\\to\\venv\\python.exe",
      "args": ["-m", "sql_mcp"],
      "env": {
        "DB_TYPE": "mssql",
        "DB_HOST": "192.168.1.50",
        "DB_USER": "sa",
        "DB_PASSWORD": "password",
        "DB_DATABASE": "DataWarehouse"
      }
    }
  }
}

🏗️ Architecture

The project uses an Adapter Pattern:

1. Server receives a request (e.g., execute_query).

2. Factory checks DB_TYPE and instantiates the correct class (e.g., MSSQLAdapter).

3. Adapter translates the generic request into specific driver calls (pymssql, pyodbc, etc.).

4. Response is formatted back to a standard JSON structure for the AI.

This means you can switch databases just by changing the config, without changing the AI's prompts or logic.

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.