Radar MCP Server

Content

Radar is a local-first Kubernetes UI from Skyhook. Its built-in MCP server lets Claude and other MCP clients query a running Radar session through Streamable HTTP, using curated tools rather than raw kubectl output.

The MCP surface is built for cluster troubleshooting and operations. It exposes token-optimized tools for dashboards, topology, neighborhoods, issues, events, logs, resource search, resource details, package inventory, Helm release inspection, RBAC review, and cluster audit findings. It also includes guarded write tools for Kubernetes resource changes, workload actions, GitOps actions, CronJob actions, and node management.

Source Review

• https://github.com/skyhook-io/radar
• https://radarhq.io
• https://github.com/skyhook-io/radar/releases
• https://raw.githubusercontent.com/skyhook-io/radar/main/README.md
• https://raw.githubusercontent.com/skyhook-io/radar/main/docs/mcp.md
• https://raw.githubusercontent.com/skyhook-io/radar/main/LICENSE
• https://raw.githubusercontent.com/skyhook-io/radar/main/internal/mcp/server.go
• https://raw.githubusercontent.com/skyhook-io/radar/main/internal/mcp/tools.go
• https://raw.githubusercontent.com/skyhook-io/radar/main/internal/mcp/tools_apply.go

These sources were reviewed on 2026-06-06. Prefer the live repository, Radar website, releases page, MCP guide, license, MCP handler source, tool registration source, and apply-tool source for current setup, transport, security controls, and tool behavior.

Features

• Built into the Radar binary and enabled by default unless Radar starts with MCP disabled.
• Streamable HTTP transport mounted on the running Radar server.
• Client setup examples for Claude Code, Claude Desktop, Cursor, Windsurf, VS Code Copilot, Cline, JetBrains AI, OpenAI Codex, and Gemini CLI.
• Read tools for dashboards, top resource usage, resource lists, resource detail, topology, neighborhoods, events, changes, pod logs, workload logs, cluster audit, package inventory, Helm releases, namespaces, search, and RBAC subject permissions.
• Diagnostic tool that bundles resource context, logs, events, and startup blockers for a workload.
• Write tools for applying YAML, patching resources, restarting, scaling, or rolling back workloads, managing CronJobs, managing ArgoCD/FluxCD resources, and cordoning, uncordoning, or draining nodes.
• Secret-data suppression, environment/log redaction, RBAC-aware access, and destructive-tool annotations in the MCP implementation.
• Local binary flow for personal cluster access and in-cluster deployment path for shared team access with authentication and RBAC planning.

Installation

Install Radar through an upstream-supported path such as Homebrew, Krew, GitHub Releases, Scoop, or the desktop installers. For Homebrew:

brew install skyhook-io/tap/radar

Start Radar with access to the intended Kubernetes context:

kubectl radar

Then configure an HTTP-capable MCP client with the Radar MCP endpoint for your chosen host and port:

{
  "mcpServers": {
    "radar": {
      "url": "RADAR_MCP_URL"
    }
  }
}

Use a local endpoint for personal development. For shared or in-cluster deployments, review Radar authentication, Kubernetes RBAC, namespace scoping, and network exposure before connecting AI clients.

Use Cases

• Ask Claude what is broken in a Kubernetes namespace before drilling into a failing workload.
• Summarize topology relationships around a Service, Deployment, Pod, or GitOps-managed application.
• Pull focused pod or workload logs after Radar has narrowed the affected resource.
• Review audit findings, RBAC permissions, package inventory, and warning events from one MCP surface.
• Investigate recent changes that may explain an incident.
• Trigger a reviewed restart, scale, rollback, CronJob run, GitOps action, or node operation only after confirming the exact scope.
• Use Radar's minified resource output when raw Kubernetes YAML would be too noisy for an LLM context window.

Safety and Privacy

Radar MCP can operate against real Kubernetes clusters. Start with the smallest namespace and RBAC scope that supports the workflow, and require human approval before any apply, patch, workload action, CronJob action, GitOps action, or node-management call. Be especially careful with server-side apply ownership, node drains, production rollbacks, and actions that interact with Helm, ArgoCD, or FluxCD state.

Treat cluster metadata as sensitive. Even with Secret data suppression and redaction controls, manifests, labels, annotations, logs, events, metrics, repository references, namespace names, RBAC rules, and topology edges can reveal customer data, security posture, incident details, or internal architecture. Keep the MCP endpoint local or authenticated, and do not expose unauthenticated in-cluster deployments to untrusted clients.

Duplicate Check

No skyhook-io/radar, Radar MCP Server, radarhq.io, or kubectl radar entry was found in content/mcp or README.md. Existing Kubernetes and Argo CD MCP entries cover different direct server implementations; the TrendRadar entry is an unrelated news and RSS monitoring project.