Stop Context-Switching Between kubectl and Your AI Assistant

You're deep in a troubleshooting session, bouncing between kubectl describe pod commands and asking Claude "what does this error mean?" There's a better way—let your AI assistant directly interact with your Kubernetes cluster.

The Context-Switching Problem

Every time you hit a K8s issue, you're doing the same dance:

1. Run kubectl commands to gather information
2. Copy/paste output to your AI chat
3. Get suggestions and run more commands
4. Repeat until resolved

This breaks your flow and wastes cognitive energy on mechanical tasks instead of solving the actual problem.

Direct AI-to-Cluster Integration

mcp-k8s-eye bridges this gap by giving your AI assistant native Kubernetes access. Instead of describing your cluster state, your AI can directly inspect, diagnose, and even fix issues.

Before:

You: "My pod is stuck in pending state"
AI: "Can you run kubectl describe pod and share the output?"
You: *runs command, copies output*
AI: "I see the issue. Can you check node resources with kubectl top nodes?"
You: *runs another command*

After:

You: "My pod is stuck in pending state"
AI: *directly inspects pod, checks node resources, analyzes events*
AI: "Found the issue - insufficient CPU on nodes. Pod needs 2 cores but max available is 1.5. Shall I scale down another deployment to free resources?"

Real Troubleshooting Scenarios

Deployment Issues: Ask "Why isn't my deployment rolling out?" and watch your AI analyze replica status, pod conditions, resource constraints, and recent events—all in one response.

Service Connectivity: Instead of manually tracing service→endpoint→pod chains, your AI can instantly map the entire connection path and identify where it breaks.

Resource Optimization: Get AI-driven recommendations based on actual usage patterns: "Your staging namespace is using 40% more memory than needed. I can optimize these 6 deployments."

Webhook Debugging: When admission controllers fail, your AI can analyze webhook configurations, check referenced services, and validate certificates—tasks that normally require multiple kubectl commands and deep K8s knowledge.

Beyond Basic kubectl

This isn't just kubectl with a chat interface. You get intelligent analysis that would take you significant time to piece together manually:

• Cross-resource diagnostics: Automatically correlate issues across pods, services, ingresses, and network policies
• Resource usage analysis: Real-time CPU/memory monitoring with optimization suggestions
• Event correlation: Connect recent cluster events to current issues
• CRD support: Full management of custom resources, not just native K8s objects

Integration That Actually Works

Two deployment modes match your workflow:

• Local stdio mode: Direct integration with tools like Cursor, Claude Desktop, or any MCP-compatible client
• SSE server mode: Deploy in-cluster for team access or remote development setups

Existing kubeconfig: Uses your current cluster access—no additional authentication setup required.

Flexible transport: Choose between stdio for local development or HTTP/SSE for shared team access.

Getting Started

# Clone and build
git clone https://github.com/wenhuwang/mcp-k8s-eye.git
cd mcp-k8s-eye
go build -o mcp-k8s-eye

# Add to your MCP client config
{
  "mcpServers": {
    "k8s-eye": {
      "command": "/path/to/mcp-k8s-eye",
      "env": {
        "HOME": "/your/home/dir"
      }
    }
  }
}

Your AI assistant now has direct cluster access. Ask it to diagnose issues, scale deployments, check resource usage, or analyze any K8s resource—without leaving your chat interface.

Transform your Kubernetes troubleshooting from a series of manual commands into natural conversation with an AI that can see and interact with your cluster directly.