GKE Cluster Creation Skill

This skill helps users create Google Kubernetes Engine (GKE) clusters by providing a set of best-practice templates and guiding them through the customization process.

core_behavior

1. Template Selection:
   • Present the available templates to the user if they haven't specified one.
   • Explain the trade-offs (e.g., Cost vs. Availability, Autopilot vs. Standard).
2. Customization:
   • Once a template is selected, present the default configuration (JSON/YAML).
   • Ask the user for essential missing information: project_id, location, cluster_name.
   • Ask if they want to modify optional fields (e.g., machineType, nodeCount, network).
3. Validation:
   • Ensure project_id, location, and cluster_name are set.
   • Ensure the configuration matches the create_cluster MCP tool schema.
4. Execution:
   • Call the create_cluster MCP tool with the final configuration.

best_practices

When guiding the user or generating configurations, adhere to the following GKE cluster creation best practices:

Security

1. Private Clusters: Default to private clusters with a private control plane and restricted public endpoints to minimize attack surface.
2. VPC-Native Networking: Use VPC-native clusters to enable alias IP ranges, which allows pod-level firewall rules and better network security.
3. Workload Identity: Prefer Workload Identity for securely granting GKE workloads access to Google Cloud services instead of using static service account keys.
4. Shielded GKE Nodes: Enable Shielded GKE Nodes to protect against rootkits and bootkits.
5. Least Privilege (RBAC): Institute strict Role-Based Access Control limits granting minimal privilege to users and workloads.

Cost Optimization

1. Autoscaling: Enable Cluster Autoscaler and Horizontal Pod Autoscaler to adjust resources based on demand.
2. Right-Sizing: Choose the appropriate machine types and node counts. Consider Spot VMs for fault-tolerant, non-critical workloads.

High Availability & Reliability

1. Regional Clusters: Use Regional Clusters for production environments to ensure control plane replication across multiple zones. (Note: standard regional creates nodes across 3 zones by default).
2. Pod Disruption Budgets: Recommend setting Pod Disruption Budgets for application stability during node maintenance.
3. Release Channels: Subscribe to a release channel (e.g., Regular or Stable) for automated and safer cluster upgrades.

templates

1. Standard Zonal (Cost-Effective Dev/Test)

Best for: Development, testing, non-critical workloads.

json
1{
2  "name": "projects/{PROJECT_ID}/locations/{ZONE}/clusters/{CLUSTER_NAME}",
3  "initialNodeCount": 1,
4  "nodeConfig": {
5    "machineType": "e2-medium",
6    "diskSizeGb": 50,
7    "oauthScopes": [
8      "https://www.googleapis.com/auth/devstorage.read_only",
9      "https://www.googleapis.com/auth/logging.write",
10      "https://www.googleapis.com/auth/monitoring",
11      "https://www.googleapis.com/auth/service.management.readonly",
12      "https://www.googleapis.com/auth/servicecontrol",
13      "https://www.googleapis.com/auth/trace.append"
14    ]
15  }
16}

2. Standard Regional (High Availability)

Best for: Production workloads requiring high availability. Note: Creates 3 nodes (one per zone in the region) by default.

json
1{
2  "name": "projects/{PROJECT_ID}/locations/{REGION}/clusters/{CLUSTER_NAME}",
3  "initialNodeCount": 1,
4  "nodeConfig": {
5    "machineType": "e2-standard-4",
6    "diskSizeGb": 100,
7    "oauthScopes": ["https://www.googleapis.com/auth/cloud-platform"]
8  }
9}

3. Autopilot (Operations-Free)

Best for: Most workloads where you don't want to manage nodes.

json
1{
2  "name": "projects/{PROJECT_ID}/locations/{REGION}/clusters/{CLUSTER_NAME}",
3  "autopilot": {
4    "enabled": true
5  }
6}

4. GPU Inference (L4)

Best for: AI/ML Inference, small model serving. Note: Requires g2-standard-4 quota.

json
1{
2  "name": "projects/{PROJECT_ID}/locations/{REGION}/clusters/{CLUSTER_NAME}",
3  "initialNodeCount": 1,
4  "nodeConfig": {
5    "machineType": "g2-standard-4",
6    "accelerators": [
7      {
8        "acceleratorCount": "1",
9        "acceleratorType": "nvidia-l4"
10      }
11    ],
12    "diskSizeGb": 100,
13    "oauthScopes": ["https://www.googleapis.com/auth/cloud-platform"]
14  }
15}

5. AI Hypercompute (A3 HighGPU)

Best for: Large Model Training/Inference. Note: High cost and strict quota requirements.

json
1{
2  "name": "projects/{PROJECT_ID}/locations/{REGION}/clusters/{CLUSTER_NAME}",
3  "initialNodeCount": 1,
4  "nodeConfig": {
5    "machineType": "a3-highgpu-8g",
6    "accelerators": [
7      {
8        "acceleratorCount": "8",
9        "acceleratorType": "nvidia-h100-80gb-hbm3"
10      }
11    ],
12    "diskSizeGb": 200,
13    "oauthScopes": ["https://www.googleapis.com/auth/cloud-platform"]
14  }
15}

instructions

• ALWAYS ask for the project_id if it is not in the context.
• ALWAYS ask for the location (Region or Zone).
• ALWAYS ask for a unique cluster_name.
• CHECK if the user wants Access to Google Cloud APIs (default cloud-platform scope is usually best for modern GKE).
• WARN the user about cost if they select GPU or Reginal clusters.
• USE create_cluster MCP tool to create the cluster. The parent argument is projects/{PROJECT_ID}/locations/{LOCATION} and the cluster argument is the JSON object. The cluster.name is just the short name (e.g. "my-cluster").
• IMPORTANT: When calling create_cluster, the cluster.name should be the short name (e.g., my-cluster), NOT the full resource path, because the parent argument defines the scope.

example_usage

User: "I want to create a GKE cluster." Model: "I can help with that. What kind of cluster do you need?

1. Standard Zonal: Good for dev/test.
2. Standard Regional: High availability.
3. Autopilot: Fully managed.
4. GPU Enabled: For AI/ML workloads."

User: "Standard Zonal, please." Model: "Great. I'll need a few details:

• Project ID
• Zone (e.g., us-central1-a)
• Cluster Name"

User: "Project my-proj, zone us-west1-b, name dev-cluster." Model: "Here is the configuration I will use: [JSON view] Do you want to proceed?"