K8M: The AI-Powered Kubernetes MCP Server That Actually Gets Things Done

Managing Kubernetes clusters while juggling AI workflows? You're probably switching between kubectl, dashboards, and AI chat interfaces constantly. K8M eliminates that context switching by being both a lightweight K8s dashboard AND a production-ready MCP server with 49+ built-in tools.

Why K8M Belongs in Your MCP Stack

Unlike traditional Kubernetes dashboards that just show you cluster state, K8M transforms your AI models into capable Kubernetes operators. When Claude or any MCP-compatible AI needs to interact with your clusters, K8M provides the secure, permission-aware bridge.

Here's what sets it apart:

• 49+ Built-in MCP Tools: Deploy pods, scale workloads, inspect logs, manage secrets - your AI can do it all through standardized MCP calls
• Permission-Aware MCP: AI operations inherit the user's actual K8s permissions - no more "AI has root access" security nightmares
• Multi-Cluster MCP: One MCP server managing multiple clusters, with fine-grained access controls per cluster
• AI-Assisted Diagnostics: Built-in Qwen2.5-Coder and DeepSeek integration for intelligent cluster analysis

Real-World MCP Integration

Instead of manually writing kubectl commands, your AI conversations become:

You: "Scale down the frontend deployment in staging to save costs"
AI: [Uses K8M MCP tools to check current replicas, then scales down]
AI: "Scaled frontend-deployment from 5 to 2 replicas in staging cluster"

You: "Why is the payment service pod crashing?"
AI: [Fetches logs, analyzes events, checks resource constraints via MCP]
AI: "Memory limit exceeded - increasing from 512Mi to 1Gi and redeploying"

The AI doesn't just tell you what to do - it actually executes the fixes through K8M's MCP interface.

Single Binary, Maximum Impact

# Download and run - that's it
./k8m --kubeconfig ~/.kube/config
# Access dashboard at localhost:3618
# MCP server available at localhost:3618/mcp

Or deploy via Docker:

services:
  k8m:
    image: registry.cn-hangzhou.aliyuncs.com/minik8m/k8m
    ports:
      - "3618:3618"
    volumes:
      - ~/.kube/config:/root/.kube/config:ro
      - ./data:/app/data

MCP Server Features That Matter

Multi-Cluster Management: Register multiple clusters, AI operations work across all of them with proper context switching.

Granular Permissions:

• Cluster-level permissions (read-only, exec, admin)
• Namespace allow/deny lists
• User and group-based access control
• MCP operations inherit the requesting user's permissions

Production-Ready Logging: Every MCP call is logged with user context, timestamp, and operation details - perfect for audit trails.

Built-in AI Models: Qwen2.5-Coder-7B and DeepSeek-R1-Distill included, or bring your own Ollama/OpenAI setup.

Integration with Your Existing Workflow

K8M doesn't replace your existing tools - it enhances them. Keep using kubectl, Helm, and your CI/CD pipelines. But now your AI assistants can:

• Troubleshoot in real-time: AI analyzes logs, events, and metrics to suggest fixes
• Automate routine tasks: Scale deployments, restart pods, update configurations
• Provide intelligent insights: Resource optimization, security recommendations, cost analysis

MCP Client Compatibility

Works with any MCP-compatible AI client:

• Claude Desktop
• Custom MCP clients
• AI agent frameworks
• Workflow automation tools

The standardized MCP protocol means you're not locked into any specific AI platform.

Getting Started with MCP

1. Deploy K8M: Single binary or Docker container
2. Configure MCP endpoint: Point your AI client to http://localhost:3618/mcp
3. Set permissions: Configure which users/groups can perform which operations
4. Start automating: Your AI can now manage Kubernetes directly

K8M transforms AI from a helpful assistant into a capable Kubernetes operator. Instead of copying and pasting kubectl commands from AI responses, your AI directly executes cluster management tasks through a secure, permission-aware interface.

Perfect for teams already using MCP who want to extend their AI capabilities into Kubernetes management - or for Kubernetes teams looking to add intelligent automation to their workflows.