Let Claude poke around your MySQL (safely)

What it does

This is a Model Context Protocol server that bridges MySQL databases to LLM tools like Claude Code, Cursor, and Codex CLI. It exposes a mysql_query tool for executing SQL, inspects schemas, and can optionally allow INSERT, UPDATE, DELETE, or even DDL operations if you flip the right environment switches. It also supports multi-database mode, SSH tunneling to remote servers, and running as a remote HTTP service with Bearer token auth.

The interesting bit

The fork leans hard into Claude Code specifically — auto-managing SSH tunnels on start/stop, importing configs from Claude Desktop, and scoping credentials to local, user, or project level. It’s essentially a database client rewritten as an LLM-readable API, with the safety rails made configurable rather than mandatory.

Key highlights

• Read-only by default; write operations require explicit ALLOW_INSERT_OPERATION, ALLOW_UPDATE_OPERATION, or ALLOW_DELETE_OPERATION flags
• SSH tunnel support with auto-start/stop hooks for remote database access
• Multi-database mode by omitting MYSQL_DB and toggling MULTI_DB_WRITE_MODE
• Remote MCP mode via HTTP with REMOTE_SECRET_KEY Bearer auth
• Connection pooling, query timeouts, rate limiting, and SSL support configurable via environment variables

Caveats

• The README is overwhelmingly setup instructions; actual behavior details are sparse beyond the tool schema
• DDL support requires disabling read-only transactions entirely (MYSQL_DISABLE_READ_ONLY_TRANSACTIONS), which feels like a big hammer for a small nail
• No mention of query result size limits or row-level security controls

Verdict

Worth a look if you want Claude Code to explore production databases without copy-pasting schemas. Skip it if you need fine-grained permission models or heavy analytics workloads — this is a convenience wrapper, not a database platform.