MCP Alchemy

What is MCP Alchemy?

MCP Alchemy connects Claude Desktop directly to your databases via SQLAlchemy, allowing it to explore database structure, write and validate SQL queries, display table relationships, and analyze large datasets. It works with PostgreSQL, MySQL, MariaDB, SQLite, Oracle, MS SQL Server, CrateDB, Vertica, and other SQLAlchemy-compatible databases. It is designed for users who want Claude to act as a database expert.

How to use MCP Alchemy?

Install uv, then add a configuration entry to claude_desktop_config.json with the command uvx, the mcp-alchemy package, and the required database driver in the --with parameter. Set the DB_URL environment variable to your database’s SQLAlchemy URL. For SQLite, no extra driver is needed; for others, include the appropriate driver (e.g., psycopg2-binary for PostgreSQL). Optionally set CLAUDE_LOCAL_FILES_PATH to enable full result set access.

Key features of MCP Alchemy

• Four built-in tools: all_table_names, filter_table_names, schema_definitions, execute_query
• Supports 10+ SQLAlchemy-compatible databases
• Connection pooling optimized for long-running MCP servers
• Integration with claude-local-files for large dataset analysis
• Smart truncation and clean vertical output for query results

Use cases of MCP Alchemy

• Explore and understand database structure without manual SQL
• Assist Claude in writing and validating SQL queries interactively
• Analyze large datasets and generate reports via claude-local-files
• Create Claude artifacts from database results for further visualization

FAQ from MCP Alchemy

What databases are supported?

PostgreSQL, MySQL, MariaDB, SQLite, Oracle, MS SQL Server, CrateDB, Vertica, and any other SQLAlchemy-compatible database.

What dependencies are required?

You need uv installed. The server runs via uvx and includes mcp-alchemy. For most databases, an additional driver package must be added with the --with parameter (e.g., pymysql for MySQL).

How does connection pooling work?

The default uses pool_pre_ping=True, pool_size=1, max_overflow=2, pool_recycle=3600, and isolation_level='AUTOCOMMIT'. These can be overridden via the DB_ENGINE_OPTIONS environment variable.

Where does data live?

Data remains in your own database. MCP Alchemy only reads and writes through the provided DB_URL connection. No data is sent to external servers.

What transport or authentication does it use?

The server communicates with Claude Desktop over stdio (standard input/output). Authentication is handled entirely by the database credentials in DB_URL.