Radar

What skyhook-io/radar MCP server does

skyhook-io/radar MCP server adds an MCP interface to Radar, an open-source Kubernetes UI and inspection tool. It lets AI agents work with cluster context covering topology, resources, events, logs, Helm state, and GitOps state. The repository also describes broader Radar views for traffic, audit information, upgrade impact, RBAC, and related operational data, but the provided material does not enumerate individual MCP tool names or schemas.

Radar is intended for environments where cluster information should remain under the operator's control. The application does not require an account, cluster agents, custom resource definitions, or a hosted cloud service. It can run locally and connect directly to Kubernetes, or it can be deployed inside a cluster for shared access.

How it works

The application runs as a single binary and uses Kubernetes API access to inspect cluster state. Local operation uses the user's Kubernetes configuration, with support for a primary kubeconfig and additional kubeconfig directories. Multiple clusters can therefore be available through the configured Kubernetes contexts. Radar also supports an in-cluster deployment through its Helm chart, where access can be limited with Kubernetes RBAC.

The skyhook-io/radar MCP server is part of the Radar runtime rather than a separately described package in the supplied documentation. Radar watches cluster state and provides live updates to its browser interface; the excerpt does not specify whether every live-update mechanism is exposed through MCP. The project states that agents can inspect, diagnose, and operate the cluster through Radar, while the exact operation set is not listed here.

Setup and configuration

Install Radar through the project's shell installer, Homebrew, Krew, Scoop, PowerShell installer, or a release download. The documented quick-start command installs Radar, after which kubectl radar launches it. An in-cluster option is available through the skyhook/radar Helm chart.

Relevant startup options include --kubeconfig, --kubeconfig-dir, --namespace, --namespaces, --namespace-scope, --port, and --listen-address. Namespace filtering can restrict the initial view, while single-namespace informer scoping is available for larger clusters. The default HTTP listener is local-only at 127.0.0.1; exposing it more broadly should be paired with authentication and network controls.

Radar can disable terminal and debug-shell access with --disable-exec, and can disable Helm write operations with --disable-helm-write. Timeline data can use memory, SQLite, or PostgreSQL storage, although those settings concern Radar's timeline feature rather than a documented MCP-specific configuration.

Tools and capabilities

The supplied material supports these capability areas:

• Query Kubernetes cluster topology and resources
• Inspect events and logs
• Examine Helm and GitOps state
• Provide cluster context to AI agents through MCP
• Work across multiple Kubernetes clusters
• Run locally or as an in-cluster deployment

The README excerpt does not provide MCP tool names, input parameters, output formats, or a client configuration example. Developers should consult Radar's MCP documentation and the installed version's --help output for those details.

Limitations and notes

The MCP interface depends on Radar's access to the target Kubernetes API and the permissions represented by the selected configuration or in-cluster identity. The excerpt does not define a separate MCP authentication scheme, transport endpoint, or supported client list. It also does not identify a standalone package command that launches only the MCP component.

Radar is Apache-2.0 licensed and described as free to use, with no account or cloud dependency required. The project supports GKE, EKS, AKS, minikube, kind, k3s, and conformant Kubernetes clusters. Some features, such as in-cluster shared access, ingress exposure, authentication, and RBAC configuration, require following the separate deployment guidance.