Kubernetes MCP Server

Add to your client

Copy the config for your MCP client and paste it into its config file.

{
  "mcpServers": {
    "kubernetes-mcp-server": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-server-kubernetes"
      ],
      "env": {
        "ALLOW_ONLY_NON_DESTRUCTIVE_TOOLS": "true"
      }
    }
  }
}

Before you start

• Node.js (it runs via `npx mcp-server-kubernetes`)
• kubectl installed and on your PATH
• Helm v3 installed if you want to use the Helm tools (optional)
• A valid kubeconfig (default ~/.kube/config) with at least one configured context pointing at a reachable cluster (minikube, GKE, EKS, Rancher Desktop, etc.)

About Kubernetes MCP Server

Tools & capabilities (11)

When to use it

• Use it when you want to inspect cluster state — pods, deployments, services, events — by asking in plain language
• Use it when debugging failing workloads (pull logs, describe pods, run the /k8s-diagnose prompt)
• Use it when deploying or upgrading apps via Helm charts or YAML manifests
• Use it when scaling deployments, managing rollouts, or port-forwarding for quick local access
• Use it when you want a safe read/patch agent — enable non-destructive mode to block deletes and teardown
• Use it when doing node maintenance like cordon/drain or cleaning up stuck pods

Quick setup

1. 1Install kubectl (and Helm v3 if needed) and confirm `kubectl get nodes` works against your cluster
2. 2Make sure your kubeconfig has the right context selected
3. 3Add the server to your MCP client, e.g. `claude mcp add kubernetes -- npx mcp-server-kubernetes` or the equivalent npx config block
4. 4Optionally set ALLOW_ONLY_NON_DESTRUCTIVE_TOOLS=true for a safer tool surface
5. 5Restart the MCP client and confirm the kubectl tools load
6. 6Verify with a read-only request such as listing pods or running the ping tool

Kubernetes MCP Server FAQ