SQLAlchemy MCP Server 🧙

A Model Context Protocol (MCP) server that enables AI assistants to interact with your databases safely and efficiently. Connect to SQLite, PostgreSQL, or MySQL databases and let AI help you explore schemas, query data, and analyze your database structure.

Note

This mcp-server is inspired by langchain's SQLDatabase-toolkit and makes use of Python's SQLAlchemy library.

What This Does

This MCP server allows AI assistants to:

• 🔍 Explore your database structure - List schemas, tables, columns, and relationships
• 📊 Query your data safely - Execute read-only queries with built-in safety controls
• 🛡️ Protect your data - Read-only mode prevents accidental data modification
• 📈 Analyze relationships - Understand how your tables connect through foreign keys
• 🔧 Support multiple databases - Works with SQLite, PostgreSQL, and MySQL

Available Tools

The server provides 8 powerful tools for database interaction:

🔍 Schema Discovery Tools

Tool	Parameters	Description	Safety
list_schemas	none	Lists all schemas in the database	✅ Safe
list_tables	schema_name (optional)	Lists all tables, optionally filtered by schema	✅ Safe
describe_table	table_name, schema_name (optional)	Shows table structure: columns, types, constraints, foreign keys	✅ Safe
get_table_relationships	none	Maps all foreign key relationships across the database	✅ Safe

📊 Data Exploration Tools

Tool	Parameters	Description	Safety
get_table_data	table_name, schema_name (optional), limit (default: 10)	Returns sample data from a table	✅ Safe
get_unique_values	table_name, column_name, schema_name (optional), limit (default: 25)	Shows unique values in a column with frequency counts	✅ Safe

⚡ Query Execution Tools

Tool	Parameters	Description	Safety
execute_read_query	sql	Executes read-only SQL (SELECT, SHOW, DESCRIBE, EXPLAIN, WITH)	✅ Safe
execute_query	sql	Executes any SQL including writes (INSERT, UPDATE, DELETE, DDL)	⚠️ Destructive

🛡️ Built-in Safety Features

• Automatic Input Validation: All table/column names are validated against SQL injection
• Result Limits: Default maximum of 25 rows returned (configurable)
• Query Timeout: Automatic timeout after 30 seconds (configurable)
• Read-Only Mode: When enabled, blocks all write operations
• Smart Query Detection: Automatically categorizes queries as safe or destructive

Tip

Reasoning models, like o4-mini, are much better at using this MCP server than regular models.

Quick Setup

Quick Reference

Database	Install Command	Dependencies
SQLite	uvx mcp-sqlalchemy	None (works out of the box)
PostgreSQL	uvx "mcp-sqlalchemy[postgresql]"	asyncpg
MySQL	uvx "mcp-sqlalchemy[mysql]"	aiomysql
All	uvx "mcp-sqlalchemy[all]"	asyncpg + aiomysql

Installation & Configuration

Option 1: Install via uvx (Recommended)

Add to your AI assistant's MCP configuration:

{
  "mcpServers": {
    "sqlalchemy": {
      "command": "uvx",
      "args": [
        "mcp-sqlalchemy[postgresql]", // Choose: [postgresql], [mysql], [all], or omit for SQLite only
      ],
      "env": {
        "DATABASE_URL": "sqlite:////absolute/path/to/database.db",
        "READ_ONLY_MODE": "true"
      }
    }
  }
}

Tip

MCP Configuration: The JSON example above shows the complete configuration needed for your AI assistant. For local development, use "command": "uv" and "args": ["run", "mcp-sqlalchemy"] instead.

Option 2: Local Development Setup

For development or customization:

# Clone the repository
git clone https://github.com/woonstadrotterdam/mcp-sqlalchemy.git
cd mcp-sqlalchemy

uv venv
source .venv/bin/activate # or .venv/Scripts/activate on Windows

# Install dependencies (choose one):
uv sync                    # SQLite only
uv sync --extra postgresql # + PostgreSQL support
uv sync --extra mysql      # + MySQL support
uv sync --extra all        # All database support

Test the Connection

# Start the development server to test using the proper entry point
uv run mcp dev src/mcp_sqlalchemy/_dev.py

This will open a web interface where you can test the connection and explore your database.

Installation Method Comparison

Method	Pros	Cons	Best For
uvx	Simple install, automatic updates, works anywhere	Requires internet, need [extra] syntax for PostgreSQL/MySQL	Most users
Local	Works offline, can modify code, full control, easy dependency management	Manual updates, manage dependencies	Developers

Updating Your Installation

For uvx installations:

# Updates happen automatically when you restart your AI assistant
# Or force update with:
uvx mcp-sqlalchemy

For local installations:

cd mcp-sqlalchemy
git pull
uv sync

Database Connection Examples

SQLite (File-based)

# Local database file
DATABASE_URL="sqlite:///database/myapp.db"

# Relative path
DATABASE_URL="sqlite:///./data/database.db"

# Absolute path
DATABASE_URL="sqlite:////full/path/to/database.db"

PostgreSQL

# Basic connection
DATABASE_URL="postgresql://user:password@localhost:5432/dbname"

# With specific schema
DATABASE_URL="postgresql://user:password@localhost:5432/dbname"
DB_SCHEMA_NAME="public"

# Remote database
DATABASE_URL="postgresql://user:password@db.example.com:5432/dbname"

MySQL

# Local MySQL
DATABASE_URL="mysql://user:password@localhost:3306/dbname"

# Remote MySQL
DATABASE_URL="mysql://user:password@mysql.example.com:3306/dbname"

Safety Features

Read-Only Mode (Recommended)

Read-only mode is enabled by default to prevent any data modifications:

# The server automatically detects and blocks:
# - INSERT, UPDATE, DELETE statements
# - DROP, CREATE, ALTER statements
# - Any potentially destructive operations

Automatic Protections

• Query Validation: Blocks dangerous SQL patterns
• Result Limits: Prevents overwhelming responses (default: 25 rows)
• Read-Only by Default: Safe mode enabled by default
• Timeout Protection: Queries timeout after 30 seconds
• Input Sanitization: Validates table and column names

What Your AI Can Do

Database Exploration

• "Show me all the tables in my database" → Lists all tables across schemas
• "Describe the users table" → Shows columns, types, and constraints
• "What tables are related to the orders table?" → Shows foreign key relationships
• "List all schemas in the database" → Shows available schemas

Data Analysis

• "Show me a sample of the users table" → Returns first 10 rows
• "What are the unique values in the status column?" → Lists distinct values with counts
• "Query all active users" → Executes: SELECT * FROM users WHERE status = 'active'
• "How many orders were placed last month?" → Custom date-based queries

Schema Understanding

• "Explain the database structure" → Comprehensive schema overview
• "How are customers connected to orders?" → Relationship mapping
• "What indexes exist on the products table?" → Index information

Configuration Options

You can customize the server behavior:

# Set maximum rows returned per query
MAX_RESULT_ROWS=50

# Set query timeout (seconds)
MAX_QUERY_TIMEOUT=60

# Enable read-only mode
READ_ONLY_MODE=true

# Set default schema (PostgreSQL)
DB_SCHEMA_NAME=public

Troubleshooting

Installation Issues

"uvx: command not found"

• Install uvx: curl -LsSf https://astral.sh/uv/install.sh | sh
• Or via pip: pip install uv then use uv tool run instead of uvx

"Module not found" for PostgreSQL/MySQL

• PostgreSQL: Install with [postgresql] syntax or ensure asyncpg is available
• MySQL: Install with [mysql] syntax or ensure aiomysql is available
• All databases: Use [all] syntax for complete support

Connection Issues

"Database URL must be provided"

• Make sure DATABASE_URL is set in your MCP configuration's env section or add --database-url to the command line
• Check the URL format matches your database type

"No module named 'greenlet'"

• For local installs: Run uv sync to install all dependencies
• For uvx installs: This should auto-install; try forcing reinstall with uvx --force ...
• The greenlet package is required for async database operations

"Connection refused"

• Verify your database server is running
• Check the hostname, port, username, and password
• For PostgreSQL/MySQL, ensure the database exists

Query Issues

"Query timeout"

• The query is taking too long (>30 seconds)
• Add LIMIT clauses to large queries
• Consider adding database indexes for better performance

"Only read-only queries allowed"

• You're in read-only mode (this is good for safety!)
• Use the "Execute Read Query" tool instead
• Or disable read-only mode if you need write access