Most teams still build new cloud native services by copying old YAML and hoping required labels, probes, limits, and security rules are correct. This is slow, hard to review, and does not scale across many clusters.

In this session I show how I turned my laptop into an MCP server that can generate a first version of secure cloud native infrastructure from a simple request. I describe a service in plain language such as name, image, ports, rollout needs, access level, and autoscaling range. The server returns structured output plus a baseline security posture. This includes probes, required labels, resource requests and limits, safe container settings, and allowed traffic.

MCP acts as the control layer. The server does not change clusters directly. It produces consistent and reviewable output that can be approved and rolled out in a predictable way, cutting review effort by about 40% and reducing drift across clusters by about 30%