skyhook-io/radar

curl -fsSL https://get.radarhq.io | sh
brew install skyhook-io/tap/radar
kubectl krew install radar
helm install radar skyhook/radar -n radar --create-namespace (in-cluster deployment)

Usage

• •kubectl radar (opens browser UI automatically at port 9280 — bare 'radar' shorthand after some installs)
• •radar cloud status (inspect an in-cluster Radar Cloud installation without modifying it)
• •Run with --no-mcp to disable the built-in MCP server, or --mcp-catalog-stdio for registry introspection over stdio

Key features

• ✓Views: Topology, Resources, Image Filesystem viewer, Timeline, Helm management, Resource Compare, TLS certs, GitOps (ArgoCD/FluxCD), Traffic, Karpenter capacity, Cost (OpenCost), Cluster Audit (31 checks), Network path diagnosis, Kubernetes upgrade impact, RBAC visibility
• ✓Built-in MCP server delivering token-optimized, pre-processed cluster data (topology graphs, health, deduplicated events, filtered logs) to AI agents; read tools are read-only, write tools run under the cluster's RBAC
• ✓Runs on your laptop against the K8s API directly — zero install on the cluster, no agents or CRDs, works on GKE/EKS/AKS/minikube/kind/k3s
• ✓Private by design: cluster data stays local, no account, no cloud sync; airgapped-friendly single binary
• ✓Real-time watches via informers with SSE updates; fast on large clusters (tested on tens of thousands of pods)

Best for

Platform engineers who want a fast, private, local-first Kubernetes dashboard that AI agents can also query through MCP.

Caveats

• ⚠Radar Cloud (hosted multi-cluster w/ SSO) is a paid product; the OSS Apache-2.0 version is fully featured and recommended
• ⚠Slow/high-latency clusters may need widened timeout flags (context-switch, first-paint, namespace-list, scope candidates)
• ⚠Auth and RBAC are off by default for local use; needed for shared in-cluster deployments