Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Kubernetes Incident Triage with k8s-mcp

    1

    Triage CrashLoopBackOff, OOM, ImagePull, Pending pods and stuck rollouts on any cluster with the read-only k8s-mcp tools.

    Free

    Security scanned
    Kubernetes Incident Triage with k8s-mcp

    Kubernetes Incident Triage with k8s-mcp

    Example session with this skill installed

    Why is the checkout pod in namespace shop crash-looping?

    • Read your context and instructions
    • Compiled the kubernetes incident triage

    Incident summary

    checkout-7d9f8c6b5-x2k4q in namespace shop (context staging) is CrashLoopBackOff. Checkout traffic in shop is failing on this replica.

    Likely root cause

    Container "app" was OOMKilled: it exceeded its 128Mi memory limit (finding pod.oom.container). The heap flag -Xmx256m is larger than the limit.

    Evidence

    • status: lastState.terminated reason=OOMKilled exitCode=137 (4m ago), from diagnose_pod
    • previous logs: java.lang.OutOfMemoryError: Java heap space, from get_logs previous=true
    • top (memory) shows the pod at the limit when metrics-server is present

    Suggested fix (for you to run)

    Set the heap well below the limit (for JVMs prefer -XX:MaxRAMPercentage=75 instead of -Xmx256m), or raise resources.limits.memory to at least 384Mi, then roll out. Do not apply this from the agent; k8s-mcp is read-only.

    Confidence

    high, if OOMKilled and the heap/limit mismatch are both present. Lower if metrics-server is missing or previous logs could not be read.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Kubernetes Incident Triage with k8s-mcp

    A companion skill for k8s-mcp, the read-only Rust MCP server for any Kubernetes cluster (EKS, GKE, AKS, k3s, kind or on-prem). It tells your agent how to install and connect k8s-mcp, confirm the cluster context, and run a symptom-led triage — then return a short incident report with evidence and suggested fix commands for you to run.

    This skill never changes cluster state and never prints Secret values.

    What it covers

    • CrashLoopBackOff, OOMKilled, ImagePullBackOff
    • Pending or unschedulable pods
    • Failing liveness, readiness or startup probes
    • Stuck rollouts
    • Node NotReady
    • Unreachable Services (endpoints, ports, DNS, LoadBalancer)

    How it works

    1. Checks whether k8s-mcp is installed and connected; if not, it gives install and MCP client config steps for Claude Code, Cursor and Claude Desktop.
    2. Confirms the kubeconfig context and guessed provider with you before switching.
    3. Starts with cluster_overview and find_problems, then the matching diagnose_* tools (diagnose_pod, diagnose_workload, diagnose_node, diagnose_service, diagnose_pvc, diagnose_dns), plus recent_changes, rollout_history, get_logs, top and autoscaling_status as needed.
    4. On EKS, GKE or AKS, uses the provider modules (eks_*, gke_*, aks_*) for node groups, add-ons, workload identity and cloud logs.
    5. Ends with an incident summary, likely root cause (with finding ids), evidence, suggested fix commands, and a confidence rating.

    Requirements

    You need k8s-mcp 0.3 or later and kubeconfig access to the cluster you want to investigate. Install with brew install riseno/tap/k8s-mcp or the release installer, then add it as an MCP server. k8s-mcp is a single static binary: no kubectl, no config file, read-only by default.

    Provider tools appear when the cluster is detected as EKS, GKE or AKS. Cloud API checks need the matching cloud credentials; without them, identity tools still return the Kubernetes side.

    This skill does not change the cluster or apply fixes. It requires k8s-mcp 0.3 or later to be installed and connected. Secret values are never returned. Provider tools (EKS, GKE, AKS) appear only when the cluster is detected and, for cloud API checks, when cloud credentials are available. Write tools are not part of k8s-mcp v0.3.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed today

    • Free with an Agensi account

    Requires

    K8s 0.3+ MCP
    kubeconfig access to a cluster
    K8s MCP

    Permissions

    Uses the k8s-mcp MCP server to read cluster state. Never writes to the cluster. Does not print Secret values.

    Listedtoday

    What's inside

    Frequently Asked Questions

    No. It is read-only. k8s-mcp has no write tools in v0.3, and the skill only suggests fix commands or YAML for you to run yourself.

    No. k8s-mcp is a single static binary with no kubectl dependency. You do need kubeconfig access to the cluster (the same file kubectl would use).

    Yes. When k8s-mcp detects those providers, this skill uses the matching modules for node groups, add-ons, workload identity and cloud logs. Other conformant clusters work with the core and diagnose tools only.

    No. This skill is a companion for riseno/k8s-mcp, a read-only Rust server that returns ranked diagnoses with evidence. It does not apply fixes.

    The skill first checks whether the MCP tools are connected. If not, it gives install and client-config steps, then continues triage once the tools appear.

    New in DevOps & Cloud