LABSCO SUMMARY

The cluster question you would have needed four CLIs to answer.

Beyond kubectl itself it reaches Helm, Flux, Argo, Istio, Velero, cert-manager, KEDA, Cilium and KubeVirt — so "why is this pod pending and did the last backup work" is one conversation instead of four terminal tabs.

What it is

A large Kubernetes control surface exposed over MCP. It uses your existing kubectl configuration, so cluster access is whatever your kubeconfig already grants.

What you get

Core objects: pods, deployments, statefulsets, daemonsets, jobs and replicasets, with get_logs, get_previous_logs, get_pod_events, check_pod_health, exec_in_pod and cleanup_pods
Diagnostics as first-class tools rather than recipes — diagnose_pod_crash, detect_pending_pods, get_evicted_pods, diagnose_network_connectivity, check_dns_resolution, trace_service_chain and compare_namespaces
Security posture: audit_rbac_permissions, check_secrets_security, get_pod_security_info and get_admission_webhooks
Cost work: get_resource_recommendations, get_idle_resources, get_cost_analysis, get_overprovisioned_resources, get_namespace_cost_allocation and optimize_resource_requests
The full Helm surface — releases, charts, repos, install, upgrade, rollback, template and chart development
Ecosystem coverage that usually needs its own CLI: GitOps with Flux and Argo, cert-manager, Kyverno and Gatekeeper policy, Velero backup and restore, KEDA, Cilium and Hubble, Argo Rollouts and Flagger, Cluster API, KubeVirt, Istio and Kiali, plus kind and vCluster
MCP resources for reading state — kubeconfig://contexts, cluster://info, cluster://nodes, namespace://list and manifest:// URIs per object type, with secret data masked
Interactive UI dashboards including show_pods_dashboard_ui, show_events_timeline_ui and show_cluster_overview_ui

Requirements

kubectl installed and configured, and access to a cluster. Either Python 3.9+ for the pip install or Node.js 14+ to run it straight through npx with no install step. A Docker image is published on both Docker Hub and the GitHub Container Registry; mount your kubeconfig read-only into the container and choose stdio or an HTTP transport. The interactive dashboards need the UI extra at install time.

Setup effort

One command — npx -y kubectl-mcp-server

See installation options

USE CASES

01

Working out why a pod keeps restarting

Diagnose the crash directly, read the previous container's logs, and check the pod's conditions and events — the three things you would otherwise fetch by hand.

diagnose_pod_crashget_previous_logsget_pod_conditionsget_pod_events

02

Chasing traffic that does not arrive

Trace the service chain, check DNS resolution inside the cluster, and diagnose connectivity, instead of shelling into a pod with curl.

trace_service_chaincheck_dns_resolutiondiagnose_network_connectivity

03

Finding what you are paying for and not using

Idle resources, overprovisioned workloads and per-namespace cost allocation, with concrete request recommendations attached.

get_idle_resourcesget_overprovisioned_resourcesget_namespace_cost_allocationoptimize_resource_requests

04

Auditing who can do what

Walk roles, cluster roles and service accounts, audit the permissions they add up to, and check how secrets are exposed.

audit_rbac_permissionsget_rbac_rolescheck_secrets_security

05

Managing Helm releases from the same conversation

List and inspect releases, read their values and manifests, then install, upgrade or roll back without switching to the Helm CLI.

helm_listhelm_get_valuesupgrade_helm_charthelm_rollback

06

Checking that backups and certificates are healthy

Velero backups and restores and cert-manager certificates and issuers are reachable as tools, including a detect call so you find out whether the component is installed at all.

backup_listbackup_detectcerts_listcerts_status_explain

251 tools

Listed in project docs. Your client may show a different list.

get_pods

get_logs

get_pod_events

check_pod_health

exec_in_pod

cleanup_pods

get_pod_conditions

get_previous_logs

get_deployments

create_deployment

scale_deployment

kubectl_rollout

restart_deployment

get_statefulsets

get_daemonsets

BEFORE YOU INSTALL

What to know first

Tools you get

251 tools named in the project's documentation.

Requires

npx on PATH
Python 3.9+
Node.js 14+
kubectl installed and configured, with access to a Kubernetes cluster

INSTALLATION

Connect your client

Server setup

TERMINAL

npx -y kubectl-mcp-server

MCP Clients

Claude Code, Cursor, Windsurf, Claude Desktop, Gemini CLI, Roo Code, Kilo Code

JSON

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

VS Code

VS CODE · JSON

{
  "servers": {
    "kubernetes": {
      "command": "npx",
      "args": [
        "-y",
        "kubectl-mcp-server"
      ]
    }
  }
}

Any MCP client — Python

ANY MCP CLIENT — PYTHON · TERMINAL

pip install kubectl-mcp-server

ANY MCP CLIENT — PYTHON · JSON

{
  "mcpServers": {
    "kubernetes": {
      "command": "python",
      "args": [
        "-m",
        "kubectl_mcp_tool.mcp_server"
      ],
      "env": {
        "KUBECONFIG": "<YOUR_PATH_TO_KUBE>/config"
      }
    }
  }
}

This runs the Python package, so install it first (Python 3.9+).

Docker MCP Toolkit

DOCKER MCP TOOLKIT · TERMINAL

docker mcp server add kubectl-mcp-server mcp/kubectl-mcp-server:latest
docker mcp server configure kubectl-mcp-server --volume "$HOME/.kube:/root/.kube:ro"
docker mcp server enable kubectl-mcp-server
docker mcp client connect claude

Kubectl MCP Server