The Kubernetes MCP runs as a **local stdio server** using your kubeconfig credentials: - **Full Cluster Access**: Uses your existing kubectl permissions—can do anything you can do - **Apply Capabilities**: Can create, modify, and delete cluster resources via `apply` - **Pod Exec**: Can execute commands inside running containers - **Credential Usage**: Leverages your local kubeconfig file for authentication **Critical security practices**: Ensure your kubeconfig has appropriate RBAC restrictions, avoid using cluster-admin contexts with this MCP, consider creating a dedicated service account with limited permissions, and be cautious with apply operations in production clusters.

To start using the Kubernetes MCP: 1. **Ensure kubectl access**: Make sure you have a working kubeconfig with cluster access 2. **Add the server**: Configure your AI client with `npx @strowk/mcp-k8s` 3. **Select context**: The MCP can list and switch between your configured contexts 4. **Start managing**: Ask your AI to inspect resources, view logs, or apply configurations **Example prompts**: - "List all pods in the production namespace" - "Show me the logs for the api-server pod" - "What events have occurred in the last hour?" - "Apply this deployment manifest to the staging cluster"