PostgreSQL DBQA with MCP (Model Context Protocol)

This project acts as a secure bridge connecting your PostgreSQL database to Artificial Intelligence (AI) models (like Claude, Cursor, etc.).

By leveraging the Model Context Protocol (MCP), it enables AI models to understand your database schema, securely query it (SELECT), and perform controlled data modifications (INSERT, UPDATE, DELETE).

Features

• Schema Analysis: Automatically introduces database tables, columns, and relationships to the AI.

• Natural Language Querying: AI translates natural language questions into SQL and retrieves results.

• Secure Write Operations:

  • Supports INSERT, UPDATE, DELETE operations.

  • Two-Phase Approval: AI first previews the affected rows using modify_data, then confirms the modification with confirm_modification.

  • Kill Switch: Write permissions can be instantly disabled with WRITE_ENABLED=false.

Related MCP server: Postgres Scout MCP

7-Layer Security Model

This server employs a multi-layered protection system to ensure data safety:

1. DDL Blocking: Structural modification commands like DROP, TRUNCATE, ALTER, CREATE are always strictly forbidden.

2. WHERE Clause Enforcement: UPDATE and DELETE queries are prevented from running without a WHERE clause.

3. Row Limit: The maximum number of rows affected by a single query is limited (Default: 100).

4. Table Whitelist: Write permission is granted only to tables specified in the .env file.

5. Two-Phase Transaction: Write operations first run in "Dry-Run" (preview) mode to calculate the number of affected rows.

6. Separate Isolation: Distinct connection managers and validators are used for read and write operations.

7. Audit Log: All operations are logged in detail.

Installation (Docker)

1. Docker image oluşturun:

   docker build -t postgres-mcp-server .

2. Bir .env dosyası oluşturun (.env.example'dan kopyalayabilirsiniz):

   copy .env.example .env   # Windows
   cp .env.example .env     # Linux/Mac

3. .env dosyasında veritabanı URI'nizi ayarlayın:

   # Bağlantı URI - tek satırda tüm bilgiler
   DATABASE_URI=postgresql://username:password@host:port/dbname
   
   # Yazma İşlemleri
   WRITE_ENABLED=true
   WRITABLE_TABLES=customers,orders,products  # Boş = tüm tablolar
   MAX_WRITE_ROWS=100

Usage

AI Client Configuration (Claude Desktop / Cursor)

Add the following to your AI client config (e.g., claude_desktop_config.json or Cursor MCP settings):

{
  "mcpServers": {
    "postgres-dbq": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm", "--network", "host",
        "-e", "DATABASE_URI",
        "-e", "WRITE_ENABLED",
        "-e", "WRITABLE_TABLES",
        "-e", "MAX_WRITE_ROWS",
        "postgres-mcp-server"
      ],
      "env": {
        "DATABASE_URI": "postgresql://postgres:your_password@localhost:5432/your_database",
        "WRITE_ENABLED": "true",
        "WRITABLE_TABLES": "",
        "MAX_WRITE_ROWS": "100"
      }
    }
  }
}

Note: --network host allows the container to access localhost services (e.g., PostgreSQL running on the host). On Windows/Mac Docker Desktop, you may also use host.docker.internal in the URI instead of localhost.

Local Development (without Docker)

pip install -r requirements.txt
python mcp_server.py

Project Structure

• mcp_server.py: Main MCP server file. Tool definitions are located here.

• src/database/:

  • executor.py: SQL execution engine (preview_write and execute_write methods).

  • schema_manager.py: Module analyzing the database schema.

• src/validation/:

  • sql_validator.py: SQL security checks and validation logic.

  • rules.py: Forbidden keywords and limit definitions.

• src/config.py: Pydantic-based configuration management.

• Dockerfile: Docker image definition.

Environment Variables

Security Warning

This tool possesses powerful capabilities. When WRITE_ENABLED=true is set, your AI model can make changes to the database.

• Configure WRITABLE_TABLES strictly before using in a Production environment.

• Ensure you have regular backups of your critical data.