PostgreSQL MCP Server

By BINATI AInalytics

An MCP (Model Context Protocol) server that provides access to PostgreSQL databases through resources, tools, and prompts for data analysis.

Demo

Features

Resources

• postgres://info - Server information and quick reference

Tools

• list_tables - List all tables in the database
• get_table_schema - Get detailed schema for a specific table
• execute_query - Execute read-only SQL queries (SELECT, WITH, SHOW)
• get_table_stats - Get statistics for a table (row count, size, indexes)

Prompts

• analyze_table - Generate a comprehensive analysis prompt for a specific table
• find_relationships - Analyze database to find relationships between tables
• data_quality_check - Perform comprehensive data quality check

Installation

This project uses uv for package management.

# Install dependencies
uv sync

Configuration

Create a .env file with your database credentials. You can either:

Option 1: Individual components (recommended)

DATABASE_HOST=localhost
DATABASE_USER=postgres
DATABASE_PASSWORD=your_password
DATABASE_PORT=5432
DATABASE_NAME=your_database

Option 2: Full connection URL

DATABASE_URL=postgresql://postgres:password@localhost:5432/mydb

Note: If you use the SQLAlchemy format postgresql+psycopg://, it will be automatically converted to the psycopg format postgresql://.

The server will automatically construct the connection URL from individual components if DATABASE_URL is not provided.

Usage

Development Mode

Test the server using the MCP Inspector:

uv run mcp dev main.py

This will launch the MCP Inspector in your browser where you can:

• Browse available resources
• Test tools with different parameters
• Try out prompts

Install to Claude Desktop

To use this server with Claude Desktop:

uv run mcp install main.py --name "PostgreSQL Server"

Direct Execution

Run the server directly:

uv run python main.py

Example Usage

Using Resources

1. Get server info:
   • Resource URI: postgres://info
   • Returns: Server information and available operations

Using Tools

1. List all tables:

   Tool: list_tables
   Returns: {"tables": [...], "count": 4}

2. Get table schema:

   {
     "table_name": "users"
   }
   Returns: {"table_name": "users", "columns": [...], "column_count": 5}

3. Execute a query:

   {
     "query": "SELECT * FROM users LIMIT 10"
   }

   Returns: JSON with rows, row_count, and columns

4. Get table statistics:

   {
     "table_name": "users"
   }

   Returns: JSON with row_count, total_size, table_size, and indexes_size

Using Prompts

1. analyze_table:

   • Generates a comprehensive analysis workflow for a specific table
   • Parameter: table_name

2. find_relationships:

   • Generates a prompt to analyze and document table relationships

3. data_quality_check:

   • Generates a prompt for comprehensive data quality analysis

Security

• Read-only queries: The execute_query tool only allows SELECT, WITH, and SHOW statements
• SQL injection protection: All queries use parameterized statements where applicable
• Connection management: Database connections are managed through lifespan context

Development

Code Quality

The project follows strict development guidelines:

# Format code
uv run --frozen ruff format .

# Check linting
uv run --frozen ruff check .

# Fix linting issues
uv run --frozen ruff check . --fix

# Type checking
uv run --frozen pyright

Testing

# Run tests
uv run --frozen pytest

Architecture

The server uses:

• FastMCP: High-level MCP server framework
• psycopg: Async PostgreSQL adapter for Python
• Lifespan management: Database connection is established at server startup and closed at shutdown
• Type safety: Full type hints throughout the codebase

License

MIT