Run kubectl tasks from plain-language requests
Ask for what you want in a terminal, such as 'show pods using the most memory', and review the kubectl commands before they run.
The use case
Engineers who touch Kubernetes only now and then forget kubectl flags and JSONPath syntax. Google's open-source kubectl-ai turns a request like 'scale the web deployment to three replicas' into kubectl commands, shows them, and runs them with confirmation, then summarizes results. It supports several model providers, including local ones. Read-only questions are low risk. For anything that changes the cluster, the engineer reads the exact command before approving, and production contexts should require confirmation every time. Teams often start by using it only for questions about cluster state, then allow changes once people trust how it reads requests.
Example workflow
- Install kubectl-ai and choose a model provider
- Ask a question or describe a task
- Review the proposed kubectl commands
- Approve to run them
- Check results and keep confirmation on for writes
Tools
- kubectl-ai
- kubectl
- Kubernetes
Risks and guardrails
- Wrong context can change the wrong cluster; check before approving
- Keep confirmation on for writes and deletes
- Cluster output sent to hosted models may include sensitive data
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Google Cloud Platform on GitHub: kubectl-ai: github.com/GoogleCloudPlatform/kubectl-ai
- Kubernetes MCP Server (official page): claude.com/marketplace/connectors/kubernetes-mcp-server (checked 2026-11-19)