MCP Server Kubernetes 🔌 MCP server Open source

What it does

MCP Server Kubernetes lets a coding agent connect to and manage a Kubernetes cluster. Written in TypeScript, it loads your kubeconfig from several sources in priority order (defaulting to ~/.kube/config) and exposes cluster operations as MCP tools. For people building software with AI, the typical uses are deploying what you just built, reading pod logs and events when a rollout fails, and letting the agent diagnose and fix a broken deployment. It wraps kubectl and Helm, so both need to be installed and on your PATH, and it can also run Helm through templates to sidestep authentication issues.

Tools it gives your agent

• A unified kubectl surface: kubectl_get, kubectl_describe, kubectl_create, kubectl_apply, kubectl_delete, kubectl_logs, kubectl_scale, kubectl_patch, kubectl_rollout, kubectl_context, explain_resource, list_api_resources and a kubectl_generic escape hatch
• Helm install, upgrade and uninstall, plus template-based apply and uninstall
• port_forward, node management (cordon, drain, uncordon), problematic-pod cleanup, and a ping connection check
• A k8s-diagnose troubleshooting prompt, secrets masking on get secrets, and optional OpenTelemetry tracing of every tool call

Works with

Any MCP client. The README documents Claude Code, Codex, Claude Desktop (including an mcpb extension), VS Code and Cursor. Runs over stdio or an SSE/HTTP transport.

How to install

claude mcp add kubernetes -- npx mcp-server-kubernetes

For Codex:

codex mcp add kubernetes -- npx mcp-server-kubernetes

Maintenance and safety

MIT-licensed and actively maintained. It can perform destructive operations, so it offers a non-destructive mode that disables deletes, uninstalls, node draining and the generic command:

ALLOW_ONLY_NON_DESTRUCTIVE_TOOLS=true npx mcp-server-kubernetes

The agent acts with your kubeconfig's permissions, so scope its credentials to what it should touch and prefer the read-only mode against production clusters.

Who should use it

Developers who deploy to Kubernetes or OpenShift and want their agent to deploy, inspect logs and troubleshoot rollouts without leaving the conversation. It assumes a working kubectl and Helm setup.