Kubernetes MCP
🔌 MCP ServerOperate and troubleshoot Kubernetes clusters in natural language
An open-source MCP server that connects to your Kubernetes cluster so the model can list pods, read logs, describe resources and run systematic troubleshooting flows. Offers a read-only/non-destructive mode with secrets masking, reuses your existing kubeconfig, and needs no extra sign-up.
Install / Add
npx mcp-server-kubernetesWhy we picked it
When your services live on Kubernetes, debugging usually means bouncing between kubectl, logs, and a pile of YAML. This MCP server hands cluster operations to the model: list pods, read logs, describe resources, run troubleshooting flows, even do a systematic diagnosis of a namespace by keyword — all in natural language. It loads kubeconfig from multiple sources in priority order, reusing the cluster credentials you already have.
It ships a non-destructive mode that allows read and create/update-only access with secrets masking, so the model can't delete production resources by accident, plus optional OpenTelemetry observability. It's MIT-licensed and connects to Claude, Cursor, Cline and others with a single npx command and no extra sign-up. For anyone living in clusters, it's a practical start to turning AI into an on-call SRE assistant.