mcp-db-server 0.1.2

MCP Database Server with read-only access

MCP Database Server

A Model Context Protocol (MCP) server implementation in Python that provides strictly read-only access to databases. It allows Large Language Models (LLMs) to inspect schemas and query data safely without risking data modification.

Features

• 🛡️ Strict Read-Only: Enforced at the driver level (e.g., SQLite ?mode=ro) or session level (MySQL SET SESSION TRANSACTION READ ONLY).
• 🏗️ Repository Pattern: Abstracted database access allows for easy extension to other backends (PostgreSQL, MySQL) via Dependency Injection.
• ⚡ FastMCP: Built efficiently using the official MCP Python SDK.
• 🔌 Connection Pooling: Automatic connection pooling for MySQL using DBUtils to ensure performance and reliability.
• 🔧 Easy Configuration: Managed via environment variables and Makefiles.

Project Structure

• mcp_db/: Main Python package for the MCP server.
  • __init__.py: Entry point and server initialization.
  • db/: Repository implementations (sqlite_repository.py, mysql_repository.py).
• seed.py: Utility script to populate the database (write access).
• Makefile: Automation for common tasks.
• pyproject.toml: Package configuration and dependencies.

Quick Start

Prerequisites

• Python 3.12+ (managed via venv)
• make (optional, but recommended)
• MySQL Server (optional, if using MySQL backend)

1. Setup

Use the Makefile to create a virtual environment, install dependencies, and seed a test database:

make setup

This runs pip install -r requirements.txt and python seed.py.

2. Configuration

The project uses python-dotenv. Copy the example config and adjust as needed:

cp .env.example .env

Key Variables:

• DB_ENGINE: sqlite (Default) or mysql
• DB_ADDRESS: Path to db file (SQLite) or Host URL (MySQL).
• DB_PORT: Database port (Default: 3306 for MySQL).
• DB_USER / DB_PASSWORD: Database credentials (required for MySQL).
• DB_SCHEMA: Database name (required for MySQL).

3. Run the Server

Start the MCP server:

make run

(Or directly via venv/bin/python server.py)

Usage with LLMs

Once running, the server exposes the following tools to connected MCP clients:

1. list_tables()
   • Returns a list of all tables in the database.
2. describe_table(table_name: str)
   • Returns the schema (columns, types, primary keys) for the specified table.
3. read_query(query: str)
   • Executes a SQL SELECT query.
   • Note: Queries attempting to modify data will raise a Security Error or OperationalError.

Development

Running Tests

Execute the unit test suite:

make test

Extending

To add support for a new database (e.g., Postgres):

1. Create db/postgres_repository.py implementation of DatabaseRepository.
2. Update server.py to instantiate it when DB_ENGINE=postgres.