K8m

What weibaohui/k8m MCP server does

weibaohui/k8m MCP server combines a web-based Kubernetes console with MCP tools for managing multiple clusters. The project describes 49 built-in Kubernetes MCP tools that can be composed into more than 100 cluster operations. These tools cover standard Kubernetes resources and custom resources defined through CRDs.

The application can register and manage multiple clusters, including standard Kubernetes, Amazon EKS, k3s, kind, and k0s environments. It can use in-cluster configuration, or scan kubeconfig files from a configured path. Cluster connections support heartbeat checks and automatic reconnection.

The dashboard provides operational functions alongside MCP access. Operators can inspect Pod logs, download logs, search log output, execute shell commands in Pods, and manage files inside Pods. CRDs are discovered automatically and displayed in a resource tree. The system also records each MCP call and connects MCP permissions with the permissions of the user making the request.

How it works

K8M is built as a Go application with an AMIS-based frontend. It uses the kom project as its Kubernetes API client. The application can run as a standalone executable, in Docker Compose, or inside a Kubernetes cluster. When deployed, it exposes a web interface on port 3618 by default; the Kubernetes deployment example uses a NodePort on 31999.

Users and groups can receive permissions per cluster. The documented permission levels are read-only access, Exec command access, and cluster administrator access. Namespace allowlists and denylists can further restrict access. This permission model also applies to MCP calls, so an agent action is evaluated using the calling user’s configured access.

The wider application includes scheduled multi-cluster inspections, custom inspection rules using Lua scripts, Kubernetes Event forwarding, webhook notifications, Helm repository management, and optional AI features. These functions are part of the K8M console and surrounding platform, rather than individually documented here as MCP tools.

Setup and configuration

Download a release and start it with ./k8m, then open http://127.0.0.1:3618. The documented default login is k8m for both username and password; the README advises changing these credentials after deployment and enabling two-step verification.

Docker Compose is also supported. The example maps port 3618, uses the registry.cn-hangzhou.aliyuncs.com/minik8m/k8m image, and mounts ./data at /app/data. Kubernetes deployment manifests are available through the repository’s deploy directory and can be applied with kubectl.

Important command-line settings include --kubeconfig, --in-cluster, --connect-cluster, --port, --login-type, and --jwt-token-secret. The executable also accepts settings for the administrator account, logging level, shell images, and image-pull timeout.

Tools and capabilities

weibaohui/k8m MCP server supports:

• Multi-cluster Kubernetes management
• Standard resource and CRD operations
• Pod log viewing and downloading
• Pod shell execution
• Pod file browsing and file transfers
• MCP call logging
• Permission-aware agent operations
• API key access for external clients

The project runs on Linux, macOS, and Windows and supports x86 and ARM architectures. It can use SQLite, MySQL, or PostgreSQL for its database. Built-in AI support includes Qwen2.5-Coder-7B, with documented support for DeepSeek-R1-Distill-Qwen-7B and user-managed models including Ollama.