Kubernetes is a Cloud Platforms MCP server that lets Claude Code, Cursor, Windsurf and any MCP-compatible AI agent cluster operations — pods, deployments, services. Install in 1 minute with mcpizy install kubernetes.
Cloud Platforms
Cluster operations — pods, deployments, services
Official homepagemcpizy install kubernetesnpx -y mcp-server-kuberneteslist_podsList pods in a namespace
Inputs
namespacestringoptionaldescribe_podDescribe a pod with events and status
Inputs
namestringrequirednamespacestringoptionalget_logsFetch logs from a pod container
Inputs
namestringrequirednamespacestringoptionallist_deploymentsList deployments in a namespace
Inputs
namespacestringoptionalscale_deploymentScale a deployment to N replicas
Inputs
namestringrequiredreplicasnumberrequirednamespacestringoptionalapply_manifestApply a YAML manifest
Inputs
manifeststringrequiredlist_nodesList all cluster nodes
Works identically across clients. Only the config file path differs.
~/.claude.json{
"mcpServers": {
"kubernetes": {
"command": "npx",
"args": [
"mcp-server-kubernetes"
]
}
}
}.cursor/mcp.json{
"mcpServers": {
"kubernetes": {
"command": "npx",
"args": [
"mcp-server-kubernetes"
]
}
}
}~/.codeium/windsurf/mcp_config.json{
"mcpServers": {
"kubernetes": {
"command": "npx",
"args": [
"mcp-server-kubernetes"
]
}
}
}Ensure ~/.kube/config is set up and your current context points at the right cluster
KUBECONFIGPaste any of these prompts into Claude Code, Cursor or another MCP-compatible client.
“List every pod in the `production` namespace and show which ones are not Running”
Uses: list_pods
“Get the last 200 log lines from the `api-server` pod”
Uses: list_pods, get_logs
“Scale `web` deployment to 5 replicas in namespace `production`”
Uses: scale_deployment
“Describe the `payments-worker` pod and tell me why it keeps restarting”
Uses: describe_pod
Kubernetes MCP exposes the Kubernetes API as agent tools: `list_pods`, `get_pod_logs`, `describe_pod`, `list_deployments`, `scale_deployment`, `apply_manifest`, `exec_in_pod`, `get_events`. Authentication is whatever the kubeconfig provides — typically a service-account token or OIDC. Most MCP forks read `~/.kube/config` by default; for cluster-scoped agents in CI, mount a service-account token instead.
We use Kubernetes MCP for three workflows. First, pod-level triage: "why is the api deployment failing" — the agent walks `list_pods` → `describe_pod` → `get_pod_logs` → `get_events` and surfaces the root cause (image pull error, crashloop, OOMKilled). Second, manifest-driven changes: the agent edits a YAML, runs `apply_manifest`, and watches `get_events` for rollout status. Third, on-call response: scaled-down replica during off-hours, the agent identifies high CPU on the remaining pod and runs `scale_deployment` to bring more online. Token cost: `describe_pod` on a verbose pod (many events, conditions) is ~3-5k tokens; `get_pod_logs` is bounded by your `tail` parameter.
Compared to Docker MCP, Kubernetes is for cluster-scale (multi-node, declarative). Docker is for single-host. Compared to Bash MCP + kubectl, Kubernetes MCP wins on structured output and on safety (kubectl arguments are easy to typo into disasters; the MCP tools are tighter). Honest trade-off: Kubernetes is complex; the MCP exposes the surface but the agent still needs to understand pods, deployments, services, ingresses, namespaces. Without solid Kubernetes context in the system prompt, the agent will produce wrong manifests.
Cluster scope is broad — a service-account token with cluster-admin can delete any resource. Scope the SA with the smallest viable RBAC role.
`apply_manifest` with a malformed YAML often returns a vague error. The agent will sometimes retry with the same broken YAML. Pin a "validate before apply" step using `kubectl --dry-run=server` equivalent if the MCP supports it.
`get_pod_logs` on a multi-container pod requires a container name; without it, some MCPs default to the first container, others error. Always pass an explicit container.
Namespace defaults trip agents. The MCP's default namespace is whatever `current-context` says — agents that forget to pass `namespace` end up querying the wrong place. Pin the namespace in every call.
Honest pros/cons against the closest cloud platforms MCP servers.
| Server | Strengths | Trade-offs |
|---|---|---|
| Docker MCP | Simpler, no cluster concepts, faster for dev workflows | Single-host only, no orchestration |
| Helm MCP (community) | Chart-level abstraction, easier for releases | Smaller surface, no raw kubectl ops |
| Bash MCP + kubectl | Full kubectl surface, every flag available | Shell injection, no structured output, typo-disaster risk |
If Kubernetes doesn't fit your stack, these Cloud Platforms MCP servers solve similar problems.
The Kubernetes MCP server is an Cloud Platforms Model Context Protocol server that lets Claude Code, Cursor, Windsurf, VS Code with Copilot, and other MCP-compatible AI agents cluster operations — pods, deployments, services. It exposes Kubernetes's capabilities as tools the AI can call directly from your editor or CLI.
The fastest way is the MCPizy CLI: run `mcpizy install kubernetes` and MCPizy will add the server to your `.claude.json` automatically. You can also install it manually by adding an entry under `mcpServers` in `.claude.json` with the command `npx -y mcp-server-kubernetes` and restarting Claude Code.
Yes. The Kubernetes MCP server is free and open source (see the GitHub repository linked on this page). You may still need a Kubernetes account or API key to connect the server to the underlying service, but the MCP layer itself has no MCPizy subscription cost.
Yes. Any MCP-compatible client works — including Claude Code, Claude Desktop, Cursor (via `.cursor/mcp.json`), Windsurf, VS Code with Copilot Chat, and custom agents built on the MCP SDK. The same install command targets all of them; only the config file path differs.
Once installed, your AI agent can cluster operations — pods, deployments, services directly inside your conversation. Typical use cases include asking Claude Code or Cursor to run Kubernetes operations, inspect results, chain Kubernetes with other MCP servers (see our Workflow Recipes), and automate repetitive cloud platforms tasks without leaving your editor.