Giving AI Agents Read-Only Database Access

I built dbridge after repeatedly hitting the same workflow gap with AI-assisted backend work: the agent could help write code, but it couldn't safely inspect the database to verify behavior.

dbridge is a cross-platform CLI + MCP server for PostgreSQL, MySQL, and MongoDB. It lets the agent query the database without exposing raw database credentials to the agent itself. Credentials stay in a dedicated OS keychain, not in config files or chat, and read-only is enforced by default.

The post explains the design trade-offs. Repo: https://github.com/thalesfp/dbridge