Kubernetes MCP Server

What kubernetes-mcp-server does

kubernetes-mcp-server exposes Kubernetes and OpenShift operations as Model Context Protocol tools. It can read kubeconfig contexts, inspect the active configuration, enumerate cluster objects, and work across multiple clusters defined in kubeconfig files. The implementation communicates directly with the Kubernetes API rather than shelling out to kubectl or helm.

Its scope covers both read and write operations. Agents can retrieve, create, update, scale, and delete arbitrary Kubernetes resources, while specialized tools handle pods, namespaces, events, nodes, Helm releases, and OpenShift projects. Additional capabilities cover kcp workspaces and Kiali or Istio service-mesh information.

How it works

The server uses the Kubernetes configuration available to the process, either from kubeconfig or an in-cluster configuration. It can detect configuration changes and update its available contexts. Requests are sent to the Kubernetes API, including API-proxy access for node logs and kubelet access for node Summary API statistics.

Many listing operations support pagination. Tools use parameters such as limit and continue, while Helm release listing uses limit and offset. Resource updates use Kubernetes Server-Side Apply: the supplied manifest represents the complete desired state, so fields omitted from an existing resource can be removed.

Setup and configuration

A Kubernetes cluster is required. The project provides a native binary for Linux, macOS, and Windows, along with npm, Python, and container or Docker distributions. Native binaries do not require Node.js or Python on the host.

For Claude Desktop, Cursor, VS Code, and Goose, the documented npm setup runs npx -y kubernetes-mcp-server@latest as the MCP command. Claude Code has a dedicated setup path, including guidance for using a dedicated read-only Kubernetes ServiceAccount in production. Configuration can be supplied through command-line arguments, TOML files, or environment variables.

Tools and capabilities

• Inspect kubeconfig contexts, server URLs, and the current configuration.
• List namespaces, OpenShift projects, events, pods, resources, nodes, and Helm releases.
• Read, create or update, scale, and delete generic Kubernetes resources.
• Retrieve pod logs, execute commands in pods, delete pods, run container images, and inspect pod metrics.
• Retrieve node logs and node, pod, or container resource statistics; node statistics can include CPU, memory, filesystem, network, and pressure metrics where supported.
• Install and uninstall Helm releases.
• List and describe kcp workspaces, and query Kiali traffic graphs, service-mesh status, and Istio-related configuration.
• Support Tekton pipeline and task operations, including starting, restarting, canceling, troubleshooting, and retrieving logs.

Limitations and notes

Operations depend on the Kubernetes or OpenShift permissions of the configured identity. Write-capable tools can modify or remove cluster resources, and Server-Side Apply requires care when a manifest omits fields from an existing object. Metrics tools rely on the corresponding Kubernetes Metrics Server or kubelet APIs. Helm listing can fail when the relevant secret list exceeds the backend response-size limit. Optional OpenTelemetry tracing and metrics are available, including a /stats endpoint for runtime statistics.