Creativity · MCP — server

MCP Kubernetes Server

The MCP Kubernetes server lets an LLM client (Claude Desktop, Cursor, Zed) interact with a Kubernetes cluster through tools that wrap kubectl and the Kubernetes API. You can ask the model to list failing pods, describe a deployment, tail logs, or apply a manifest — all scoped to a kubeconfig context. Most implementations are community-maintained rather than official Anthropic reference servers, so verify the repo before production use.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: TypeScript / Node.js
Transports: stdio

Capabilities

Tools: list_pods, describe_resource, get_logs, apply_manifest, delete_resource
Resources: cluster, namespace, and pod metadata as MCP resources
Safety: operations run with the permissions of the provided kubeconfig

Install

npx -y mcp-server-kubernetes

Configuration

{
  "mcpServers": {
    "kubernetes": {
      "command": "npx",
      "args": ["-y", "mcp-server-kubernetes"],
      "env": {
        "KUBECONFIG": "/Users/you/.kube/config"
      }
    }
  }
}

Frequently asked questions

Is there an official MCP Kubernetes server from Anthropic?

Not as of April 2026. Several community implementations exist (e.g. mcp-server-kubernetes on npm, Go-based variants on GitHub). Check the maintainer, release date, and open issues before using it against a production cluster.

What permissions does the MCP Kubernetes server need?

It reuses your kubeconfig, so it inherits whatever RBAC your user has in that context. For production, create a dedicated ServiceAccount with least-privilege RoleBindings rather than pointing it at cluster-admin.

Can the model delete resources?

Yes, if the server exposes a delete_resource tool. Most Claude clients require user approval per tool call, so destructive actions prompt before executing — but you should still run against non-prod clusters first.

Sources

Model Context Protocol — servers directory — accessed 2026-04-20
Kubernetes API reference — accessed 2026-04-20