Kubernetes MCP Server

An MCP server that provides tools for generating Kubernetes YAML manifests and deploying using kubectl.

Features

Manifest Generation

• generate_deployment: Create Deployment manifests with resource limits, env vars
• generate_service: Create Service manifests (ClusterIP, NodePort, LoadBalancer)
• generate_configmap: Create ConfigMap manifests
• generate_secret: Create Secret manifests (auto base64 encoding)
• generate_ingress: Create Ingress manifests with optional TLS
• generate_namespace: Create Namespace manifests

Kubectl Operations

• kubectl_apply: Apply manifests to cluster
• kubectl_delete: Delete resources from cluster
• kubectl_get: List resources
• kubectl_describe: Describe resources in detail
• kubectl_logs: Get pod logs
• kubectl_exec: Execute commands in pods

Prerequisites

• Python 3.10+
• kubectl installed and configured
• Access to a Kubernetes cluster (kubeconfig set up)

Installation

cd kubernetes-mcp-server
pip install -r requirements.txt

Kubernetes Access

Ensure kubectl is configured:

# Verify connection
kubectl cluster-info

# Check current context
kubectl config current-context

# List available contexts
kubectl config get-contexts

Configuration

Add to your Claude Code settings (~/.claude/settings.json):

{
  "mcpServers": {
    "kubernetes": {
      "command": "python",
      "args": ["/Users/youruser/Documents/Base/DevOps-ClaudeAi/test-cases/SM1/my-first-vpc/kubernetes-mcp-server/server.py"]
    }
  }
}

Usage Examples

Generate Manifests

User: "Generate a deployment for nginx with 3 replicas"
Claude: [calls generate_deployment with name=nginx, image=nginx:latest, replicas=3]

User: "Create a LoadBalancer service for the nginx deployment"
Claude: [calls generate_service with service_type=LoadBalancer, selector={app: nginx}]

User: "Generate a configmap with database connection settings"
Claude: [calls generate_configmap with data]

Deploy to Cluster

User: "Apply the nginx deployment to the cluster"
Claude: [calls kubectl_apply with manifest path]

User: "Get all pods in the default namespace"
Claude: [calls kubectl_get with resource_type=pods, namespace=default]

User: "Show me the logs from the nginx pod"
Claude: [calls kubectl_logs with pod_name]

Manifest Storage

Generated manifests are automatically saved to ~/k8s-manifests/ with timestamps:

• deployment-nginx_20260417_143022.yaml
• service-nginx_20260417_143045.yaml

This allows you to:

• Track generated configurations
• Version control your manifests
• Re-apply manifests later
• Review what was created

Security Notes

⚠️ IMPORTANT:

• This server can deploy resources to your Kubernetes cluster
• Always review generated manifests before applying
• Use RBAC to limit kubectl permissions
• Be cautious with kubectl_exec (can run arbitrary commands in pods)
• Secrets are base64 encoded (not encrypted) - use proper secret management solutions

Recommended RBAC Policy

Create a ServiceAccount with limited permissions:

apiVersion: v1
kind: ServiceAccount
metadata:
  name: mcp-deployer
  namespace: default
---
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  name: mcp-deployer-role
  namespace: default
rules:
- apiGroups: ["", "apps", "networking.k8s.io"]
  resources: ["deployments", "services", "configmaps", "secrets", "ingresses"]
  verbs: ["get", "list", "create", "update", "patch"]
- apiGroups: [""]
  resources: ["pods", "pods/log"]
  verbs: ["get", "list"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: mcp-deployer-binding
  namespace: default
subjects:
- kind: ServiceAccount
  name: mcp-deployer
roleRef:
  kind: Role
  name: mcp-deployer-role
  apiGroup: rbac.authorization.k8s.io

Working with Multiple Clusters

Use the context parameter to target different clusters:

User: "Deploy to production cluster"
Claude: [calls kubectl_apply with context=prod-cluster]

User: "List pods in staging"
Claude: [calls kubectl_get with context=staging-cluster]

Error Handling

The server provides detailed kubectl error messages:

• Resource already exists
• Permission denied (RBAC)
• Invalid manifest syntax
• Connection errors
• Resource not found