googleapis/mcp-toolbox — reverse-engineered prompt
Reverse engineered prompt
Build me an open source MCP server that lets AI tools like Claude Code, Codex, Gemini CLI, and other MCP clients safely connect to databases.
I want someone to be able to point it at a database and immediately ask for things like list my tables, inspect the schema, run a query, and help generate database aware code. It should also let more advanced users define their own safe tools in a simple YAML config, with fixed queries, parameters, semantic search, and controlled access so agents aren’t just doing anything they want.
Support common databases like Postgres, MySQL, SQL Server, BigQuery, Cloud SQL, Spanner, Firestore, MongoDB, Redis, Elasticsearch, Snowflake, Oracle, and similar systems. Include sensible connection handling, authentication options, logging, metrics, and tracing.
Make it runnable from the command line, Docker, and as an MCP server over stdio. Include clear docs, examples, and simple SDK style ways for apps to load and call the tools. Look up current MCP docs online if needed.
Want more depth? Deep Reverse