k8s_mcp_server_prod by samcolon - MCP Server

🧠 AI-Powered Kubernetes MCP Server (Minikube + Gemini + kubectl-ai)

This project deploys an AI-connected Kubernetes MCP (Model-Controller-Prompt) server on an EC2 instance using Minikube, Google Gemini, and kubectl-ai. The MCP server allows natural-language interaction with Kubernetes — powered by a custom schema and FastAPI backend.

📖 Full Guide Available

📝 Check out the full Medium article walkthrough here:
👉 Read on Medium

🧱 Architecture Overview

EC2 Ubuntu Instance (T2.Medium, 30 GiB EBS)
Minikube (Docker driver)
kubectl-ai – Google’s CLI for AI-driven Kubernetes commands
Gemini API (1.5 Flash) – Natural language LLM
FastAPI MCP Server – Hosts custom mcp-schema.json for command interpretation
NodePort Service – Exposes MCP server endpoint
Demo App – my-website-app deployed for live testing

📦 Prerequisites

Tool	Required
Ubuntu 22.04 EC2	✅
Docker (non-root)	✅
Minikube	✅
Go 1.22+	✅
kubectl	✅
Gemini API Key	✅

⚙️ Setup Steps

🔐 1. Create Security Group

Allow SSH (22)
Allow NodePort range: 30000–32767 (TCP)
Allow HTTP (80)

☁️ 2. Launch EC2 Instance

AMI: Ubuntu 22.04+
Type: t2.medium or higher
Disk: 30 GiB
User-data:
```
#!/bin/bash
```

set -e

Update system and install essentials

apt-get update -y && apt-get upgrade -y apt-get install -y curl wget git ca-certificates gnupg lsb-release apt-transport-https software-properties-common

---- Python ----

apt-get install -y python3 python3-pip update-alternatives --install /usr/bin/python python /usr/bin/python3 1 update-alternatives --install /usr/bin/pip pip /usr/bin/pip3 1

---- Docker ----

apt-get install -y docker.io systemctl enable docker systemctl start dockeru


---

### 🧑‍💻 3. Initial Setup After SSH

```bash
sudo usermod -aG docker ubuntu
sudo reboot

📦 4. Install Requirements

# kubectl

# Go

Follow official Minikube install guide:
👉 https://minikube.sigs.k8s.io/docs/start/

📁 5. Clone Repos

git clone https://github.com/your-username/k8s-mcp-server-prod.git
cd k8s-mcp-server-prod

git clone https://github.com/GoogleCloudPlatform/kubectl-ai.git
cd kubectl-ai
go build -o kubectl-ai ./cmd/kubectl-ai
sudo mv kubectl-ai /usr/local/bin/

☸️ 6. Start Minikube

minikube start --driver=docker

🔌 7. Deploy MCP + App

kubectl apply -f rbac.yaml
kubectl apply -f my-website-app.yaml
kubectl apply -f mcp-deployment.yaml
kubectl apply -f mcp-service.yaml

🔍 8. Test MCP Server

minikube ip  # e.g. 192.168.49.2
kubectl get svc mcp-service  # Note NodePort, e.g. 31390

curl http://192.168.49.2:31390/mcp-schema.json

✅ Should return your schema in JSON.

📘 9. Configure `kubectl-ai`

mkdir -p ~/.kube/kubectl-ai
nano ~/.kube/kubectl-ai/config.yaml

Paste:

mcp:
  endpoint: http://192.168.49.2:31390/mcp-schema.json
  name: mcp-server

llm:
  provider: gemini
  model: gemini-1.5-flash

🔑 10. Create Gemini Secret

kubectl create secret generic gemini-api-key   --from-literal=GEMINI_API_KEY=your-key

export GEMINI_API_KEY=your-key
echo 'export GEMINI_API_KEY=your-key' >> ~/.bashrc
source ~/.bashrc

🧠 11. Use `kubectl ai`

kubectl ai --model gemini-1.5-flash

💬 Demo Prompts

You can now ask kubectl ai things like:

Prompt	Action
List all pods in the default namespace	`list_pods`
Restart the my-website-app deployment	`restart_deployment`
Scale the my-website-app to 5 replicas	`scale_deployment`
Delete pod `my-website-app-xyz`	`delete_pod`
Get logs from pod in default namespace	`get_pod_logs`
List all nodes	`get_nodes`
Get cluster namespaces	`get_namespaces`
Get events in default namespace	`get_events`

🧹 Cleanup

kubectl delete -f mcp-service.yaml
kubectl delete -f mcp-deployment.yaml
kubectl delete -f my-website-app.yaml
kubectl delete -f rbac.yaml
kubectl delete secret gemini-api-key

🙌 Acknowledgements

📌 Next Steps

Add HTTPS ingress controller with TLS
Package MCP as a Helm chart
Add multi-model support for OpenAI or Anthropic