MCP K8s Eye

What wenhuwang/mcp-k8s-eye MCP server does

The wenhuwang/mcp-k8s-eye MCP server exposes Kubernetes management and analysis functions through MCP. It can work with standard Kubernetes objects such as Pods, Deployments, Services, StatefulSets, and Ingresses, as well as CustomResourceDefinition-backed resources. Resource operations include listing, retrieving, creating or updating, deleting, and describing objects within a namespace.

It also provides focused operational actions. An agent can retrieve pod logs, execute commands inside a pod, and scale a Deployment. These operations make the server suitable for workflows that combine inspection with changes to an application environment.

How it works

The server connects to a Kubernetes cluster using the configuration available to kubectl. The README requires kubectl to be configured before use, so cluster access and credentials are supplied through the local Kubernetes setup rather than through an MCP-specific API key.

Two transport modes are documented. Stdio runs the compiled server as a local process and can receive a HOME value to identify the directory containing the kubeconfig. SSE uses an HTTP endpoint, shown in the configuration as http://localhost:8080/sse, after an SSE server has been started.

The wenhuwang/mcp-k8s-eye MCP server includes analysis tools that inspect object status and related configuration. Workload monitoring reports CPU and memory usage for Pods, Deployments, ReplicaSets, and StatefulSets, while node analysis checks node conditions.

Setup and configuration

Building the wenhuwang/mcp-k8s-eye MCP server requires Go 1.23 or newer. The documented setup is to clone the repository, enter its directory, and build a binary named mcp-k8s-eye with Go. The Stdio client configuration must then point its command value to the resulting binary; the README uses a placeholder path, so the actual path depends on the local checkout.

For Stdio use, set HOME to the user home directory when that is needed to locate the kubeconfig. For SSE use, configure the client with the server's actual SSE URL. The example uses an endpoint on localhost and does not document authentication settings.

Tools and capabilities

Resource tools include resource_get, resource_list, resource_create_or_update, resource_delete, and resource_describe. Operational tools include deployment_scale, pod_exec, and pod_logs.

Analysis tools cover Pods, Deployments, StatefulSets, Services, CronJobs, Ingresses, NetworkPolicies, ValidatingWebhooks, MutatingWebhooks, and Nodes. The monitoring tool, workload_resource_usage, reports CPU and memory usage for supported workload types.

Limitations and notes

Cluster-wide diagnostics and cluster-level CPU or memory capacity analysis are listed as unsupported. Resource explanation is also not implemented. The documented monitoring scope does not include node or cluster utilization, even though node condition analysis is available.

The README states that the server supports multiple AI clients and shows a Cursor tools screenshot, but it does not provide a complete client compatibility matrix. Kubernetes permissions, namespace access, and the effect of mutating operations depend on the configured cluster credentials and RBAC rules.