Your AI Doesn't Need to Dig Around in Production Databases. It Needs Better Briefings.
Category:
sql-server
, postgres
I increasingly hear or read the advice to solve problems by querying MCP servers on production databases.
Meanwhile, I’m reading output from an AI agent that tried to use an MCP server of some sort. And the output is full of things that are obviously wrong/false/dangerous.
It’s expensive and dangerous to trust an AI Agent to query production freely, even with read-only access. You have MUCH safer options which will give you better results, both for human employees and AI Agents.