Agentic-Kubernetes-CLI

The MCP server directly executes 'kubectl' commands generated by the AI agent based on user input. While 'shlex.split' is used to mitigate basic shell injection, the 'validate_kubectl_command' function in 'k8s_cli_mcp_server.py' explicitly allows dangerous 'delete', 'drain', and 'cordon' operations to 'pass' without actual validation or user confirmation by the tool itself. The system relies on the LLM's prompt to 'confirm the command with the user first' for destructive operations, which is not a robust security mechanism. A confused or malicious LLM could issue highly destructive commands without sufficient guardrails, posing a significant risk to the Kubernetes cluster.