MCP Server for Database (PostgreSQL and MySQL)

A Model Context Protocol server that provides access to PostgreSQL or MySQL. This server enables LLMs to inspect database schemas and execute queries.

Key Features

• Autonomous Schema Discovery: AI agents can list tables and inspect columns to understand your data structure without manual intervention.
• Security First (Read-Only): Strictly enforces read-only access. Only SELECT, WITH, SHOW, DESCRIBE, and EXPLAIN statements are permitted.
• Multi-Database Support: Seamlessly connects to PostgreSQL and MySQL (SQL Server & SQLite support coming soon).
• Flexible Deployment: Supports multiple transport modes including stdio (for local IDEs), HTTP/SSE (for cloud APIs), and Docker.

Available Tools

The server exposes the following tools to AI clients:

Tool	Capability	User Intent Example
list-tables	Lists all accessible tables in the database.	"What tables are available in the schema?"
describe-table	Retrieves column details, data types, and constraints.	"Tell me about the 'Inventory' table structure."
execute-query	Executes safe, read-only SQL queries to fetch live data.	"Who are the top 5 customers by revenue today?"

Architecture Overview

graph TD
    A[AI Assistant] -->|MCP Protocol| B{MCP Server}
    subgraph "Deployment Modes"
        B -->|stdio| C[Local IDE: VSCode/Cursor]
        B -->|HTTP/SSE| D[Cloud/Remote API]
        B -->|Docker| E[Containerized Service]
    end
    B -->|Read-Only SQL| F[(PostgreSQL / MySQL)]

[!TIP] Dive Deeper: For a detailed breakdown of the communication flow, sequence diagrams, and deployment strategies, please check out our Full Architecture Guide

Quick Start

1. Installation

pip install -r requirements.txt

2. Run with Docker (Recommended)

Update the DATABASE_URI in docker-compose.yaml and run:

docker compose up --build

3. Run Locally

# PostgreSQL
python -m src --connection-string "postgresql://user:pass@localhost:5432/db"

# MySQL
python -m src --connection-string "mysql+pymysql://user:pass@localhost:3306/db"

AI Client Configuration

Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "sql-intelligence": {
      "command": "python",
      "args": ["-m", "src", "--connection-string", "your_connection_string"],
      "cwd": "/path/to/project"
    }
  }
}

VSCode (Cline / Roo Code)

Create .vscode/mcp.json in your workspace:

{
  "mcpServers": {
    "database": {
      "command": "python",
      "args": [
        "-m", "src",
        "--connection-string", "postgresql://user:pass@localhost:5432/mydb"
      ],
      "cwd": "/path/to/mcp-for-database"
    }
  }
}

Security Guardrails

The server utilizes a strict Regex-based validation layer. Any query that does not start with an authorized read-only keyword is blocked immediately to prevent unauthorized data modification.

Roadmap

• [ ] Support for SQL Server and SQLite.
• [ ] Integration of Audit Logs for tracking AI query behaviors.
• [ ] Semantic Search capabilities combining SQL and Vector DB.

Project Demo

You can watch the full demo of this MCP Server for Database on YouTube: Watch here