AI + MCP Integration Clients¶
This directory contains Go-based clients to integrate AI assistants with your Kubernetes PDB Management MCP server.
Available Clients¶
1. Claude Client (claude-mcp)¶
Uses Anthropic’s Claude API with MCP tools for Kubernetes cluster analysis.
2. ChatGPT Client (chatgpt-mcp)¶
Uses OpenAI’s ChatGPT API with MCP tools for Kubernetes cluster analysis.
Prerequisites¶
Port-forward running: MCP server accessible at localhost:8090
kubectl port-forward -n canvas svc/pdb-management-mcp-service 8090:8090
API Keys: Get API keys from respective providers:
Claude: Anthropic Console
ChatGPT: OpenAI Platform
Setup¶
Claude Client Setup¶
Set API key:
export ANTHROPIC_API_KEY='sk-ant-...' # Your actual key
Build and run:
go build -o claude-mcp claude_mcp_client.go ./claude-mcp
ChatGPT Client Setup¶
Set API key:
export OPENAI_API_KEY='sk-...' # Your actual key
Build and run:
go build -o chatgpt-mcp chatgpt_mcp_client.go ./chatgpt-mcp
Usage Examples¶
Both clients support the same types of queries:
Interactive Commands¶
> analyze pdb-demo
> What's the PDB coverage in canvas namespace?
> Which deployments are most vulnerable to node failures?
> Create availability policies for payment services in pdb-demo namespace
> Compare availability between pdb-demo and canvas namespaces
> quit
As a Library¶
Claude Integration:
integration, err := NewClaudeMCPIntegration(apiKey, "http://localhost:8090/mcp")
if err != nil {
log.Fatal(err)
}
ctx := context.Background()
response, err := integration.AnalyzeNamespace(ctx, "pdb-demo")
fmt.Println(response)
ChatGPT Integration:
integration, err := NewChatGPTMCPIntegration(apiKey, "http://localhost:8090/mcp")
if err != nil {
log.Fatal(err)
}
ctx := context.Background()
response, err := integration.AnalyzeNamespace(ctx, "pdb-demo")
fmt.Println(response)
Available MCP Tools¶
Both clients have access to all 7 MCP tools:
analyze_cluster_availability - Analyze cluster-wide PDB coverage
analyze_workload_patterns - Identify deployment patterns
recommend_availability_policies - Get policy recommendations
create_availability_policy - Create availability policies
update_deployment_annotations - Update deployment annotations
simulate_policy_impact - Simulate policy changes
get_deployment_availability_status - Get deployment status
Key Features¶
Native Go implementation - Single binary deployment
Pretty JSON formatting - Human-readable output
Tool calling support - Full MCP protocol implementation
Interactive CLI mode - Easy to use interface
Library integration - Can be embedded in other Go applications
Error handling - Comprehensive error reporting
Timeout management - Configurable timeouts
Model Comparison¶
Feature |
Claude Client |
ChatGPT Client |
|---|---|---|
Model |
|
|
Max Tokens |
1024 |
1024 |
Tool Support |
✅ Native |
✅ Function Calling |
Cost |
~$0.25/1M input tokens |
~$10/1M input tokens |
Speed |
Very Fast |
Fast |
JSON Output |
Excellent |
Excellent |
Troubleshooting¶
Connection Issues¶
Port-forward not running: Start with
kubectl port-forwardMCP server unhealthy: Check with
curl http://localhost:8090/healthAPI key issues: Verify environment variables are set
API Errors¶
Claude: Check ANTHROPIC_API_KEY is valid
ChatGPT: Check OPENAI_API_KEY is valid and has credits
Performance Tips¶
Use Claude for cost-effective analysis
Use ChatGPT for more detailed explanations
Both support the same MCP tools and capabilities
Example Session¶
$ ./claude-mcp
Initializing Claude + MCP integration...
MCP server initialized successfully
Loaded 7 MCP tools
Claude + MCP Integration Ready!
--------------------------------------------------
> Which deployments would be affected by a node failure?
Claude is calling tool: analyze_cluster_availability
Tool result:
{
"summary": {
"totalDeployments": 23,
"deploymentsWithPDB": 17,
"deploymentsWithoutPDB": 6,
"coveragePercentage": 73.91
},
"issues": [
{
"type": "deployment",
"severity": "warning",
"resource": "coredns",
"namespace": "kube-system",
"description": "Multi-replica deployment without PDB"
}
]
}
Claude: Based on the analysis, 6 deployments would be most affected by node failures:
1. **coredns** (kube-system) - 2 replicas, no PDB - CRITICAL
2. **cert-manager** components - Single replica each, no PDBs
3. **metrics-server** (kube-system) - Single replica, no PDB
...
Integration Benefits¶
✅ Natural Language Interface - Ask questions in plain English
✅ Real-time Cluster Analysis - Live data from your Kubernetes cluster
✅ Intelligent Recommendations - AI-powered policy suggestions
✅ Automated Policy Creation - Generate policies from descriptions
✅ Multi-namespace Analysis - Compare across namespaces
✅ Rich Formatting - Clean, readable JSON output