kubeview-mcp: Skill for Claude Code

Kubernetes Cluster Debugging

Reusable investigation playbooks for debugging Kubernetes clusters using KubeView MCP.

Core Principles

• Read-Only: Avoid state changes unless explicitly authorized.
• Prefer MCP Tools: Use kube_* tools over kubectl commands.
• Security First: Treat output as potentially sensitive.
• Declarative Fixes: Provide YAML snippets rather than imperative kubectl commands.

Tool Quick Reference

• Default surface: run_code, approval-gated kube_pod_exec
• Code discovery: tools.list(), tools.search(), tools.help(), tools.disabled()
• Kubernetes namespace: list/get resources, metrics, unified logs, network probes, and configurable port-forward
• Direct tools mode: kube_list, kube_get, kube_logs, kube_contexts, helm, kube_pod_exec, plus detected argo and argocd
• Other clusters: list targets with kube_contexts (tools.kubernetes.contexts()), then pass context (name or alias) on any call, or use tools.withContext('staging') / the run_code context argument

In default code mode, capabilities use typed namespaces such as tools.kubernetes.list and tools.helm.get. When the user names a cluster or environment ("in staging"), pass it as context rather than assuming the default context.

Debugging Decision Tree

Issue reported
    │
    ├─ Pod not running? ──────────► Skill: Debug Pod Failures
    │
    ├─ Service unreachable? ──────► Skill: Debug Service Connectivity
    │
    ├─ Deployment stuck? ─────────► Skill: Debug Deployment Rollout
    │
    ├─ Node issues? ──────────────► Skill: Node Debugging
    │
    └─ Performance/Resources? ────► Skill: Resource Debugging

---

Skill: Cluster Triage

When: Get high-level cluster health overview, identify problem areas.

Trigger: "cluster health", "triage the cluster", "what's wrong"

Steps:

1. Cluster diagnostics

   { "tool": "kube_list", "args": {} }
   

2. Metrics + top consumers