K8sGPT

K8sGPT is open-source Kubernetes diagnosis with an MCP server for AI assistants.

K8sGPT runs targeted analyzers across pods, services, deployments, ingresses, and events, then asks an LLM (OpenAI, Azure, Bedrock, or local) to explain failures in plain English. Operator mode runs the analyzers continuously inside the cluster, and the MCP server exposes those analyzers as tools so Claude Code or other agents can investigate cluster issues as part of a broader workflow. Experimental auto-remediation patches a narrow set of resources.

Who it's for. Platform engineers and SREs of 1 to 15 people who want AI-assisted Kubernetes debugging without paying for a commercial SRE platform. The example: a deployment fails with an opaque ImagePullBackOff, k8sgpt analyze names the missing tag and points at the CI pipeline, and Claude Code can run the same analysis as part of a longer troubleshooting session via MCP.

Tradeoffs. Diagnosis only; no incident management, alert correlation, or service catalog. Auto-remediation is experimental and should not run in production unsupervised. Requires an external AI backend, which adds API cost and data egress concerns for sensitive clusters. Free and open source.

Compare: Komodor, Robusta, Botkube, Datadog Bits AI