Summary
It inherits your kubeconfig, so it reaches exactly what you can reach.
Because everything runs through kubectl, the cluster it touches is the one your current context points at — and switching contexts is itself one of the operations, which is worth remembering before a scale command. Cluster roles, role bindings, namespaces and service accounts are listed as upcoming, not shipped.
What it is
An MCP server for Kubernetes that wraps kubectl, so an assistant can inspect and change resources in whatever cluster your current context points at.
What you get
- Create, update, scale and delete deployments in a namespace
- List pods, namespaces, nodes, services, deployments, jobs, cronjobs, statefulsets and daemonsets
- Read pod logs and namespace events
- Add or remove labels and annotations on a pod
- Expose a deployment on a port, and port-forward a pod, deployment or service to a local port
- Show the current context, list contexts and switch between them
Requirements
Cluster access already configured through kubectl, plus Python 3.x. Your client runs kubernetes.py through uv from the project directory; a Smithery command installs it for Claude Desktop.
Setup effort
One command — npx -y @smithery/cli install @abhijeetka/mcp-k8s-server --client claude