What happens when you hand an AI agent the keys to everything, on purpose? I built the worst case into my real homelab, then dismantled it: just-in-time access, tap-to-approve, and the standing creds I am still honest about keeping.
Every alert spends a human's attention budget. Design your monitoring as if that budget were the scarce resource it actually is, because it is.
Guardrails won't save you. AI agents need real access control, lifecycle management, and identity governance — the same way any employee does.
I built a single MCP server that gives Claude Code full operational control over my homelab — 112 tools across twelve services, from Kubernetes and Terraform to secret management and monitoring. Here's how MCP turns an AI assistant into an infrastructure operator.
You lock down your firewall and rotate your secrets, then blindly helm install a chart from a repo with 12 stars. Supply chain risk is the homelab blind spot nobody talks about.