OpenShift OVN-Kubernetes Benchmark MCP Server

# OpenShift OVN-K Benchmark MCP Client with LangGraph This project provides an AI-powered chat interface for analyzing OpenShift OVN-Kubernetes performance using MCP (Model Context Protocol) and LangGraph for intelligent agent orchestration. ## Architecture - **MCP Server**: `ovnk_benchmark_mcp_server_detailed_info.py` - Provides tools for cluster analysis - **MCP Client**: `ovnk_benchmark_mcp_client_chat.py` - LangGraph-powered chat interface with memory - **Web Interface**: `ovnk_benchmark_mcp_llm.html` - Interactive web UI for chat and data visualization ## Features - **Intelligent Agent**: LangGraph-powered React agent with persistent memory - **MCP Integration**: Streamable HTTP connection to MCP server tools - **Real-time Chat**: Streaming responses with SSE (Server-Sent Events) - **Data Visualization**: Automatic table formatting for structured data - **Memory Persistence**: Conversation memory across sessions - **Tool Auto-discovery**: Agent automatically learns available tools and parameters ## Prerequisites 1. **OpenShift Cluster**: Running OVN-Kubernetes networking 2. **Prometheus**: Accessible for metrics collection 3. **Python 3.8+**: Required for both server and client 4. **OpenAI API Key**: For LLM functionality (or configure alternative LLM) ## Installation 1. **Install Dependencies**: ```bash pip install -r requirements.txt ``` 2. **Set Environment Variables**: ```bash export OPENAI_API_KEY="your-openai-api-key" # Or configure alternative LLM provider ``` 3. **Configure OpenShift Access**: Ensure you have proper kubeconfig and Prometheus access configured. ## Usage ### 1. Start the MCP Server First, start the MCP server that provides the analysis tools: ```bash python ovnk_benchmark_mcp_server_detailed_info.py ``` The server will start on `http://localhost:8000` and provide the following tools: - Cluster health checks and general information - Node resource usage analysis - API server performance metrics - OVN-K pod and container metrics - OVS performance analysis - Synchronization duration metrics - Comprehensive performance analysis ### 2. Start the MCP Client In a new terminal, start the LangGraph-powered client: ```bash python ovnk_benchmark_mcp_client_chat.py ``` The client will start on `http://localhost:8001` and provide: - `/api/health` - Health check endpoint - `/chat/stream` - Streaming chat interface - `/api/tools/{tool_name}` - Direct tool invocation ### 3. Open the Web Interface Open `ovnk_benchmark_mcp_llm.html` in your browser. The interface will connect to the client on port 8001. ## Available Commands The AI agent can understand natural language requests and automatically call appropriate tools: ### Quick Actions - **"Perform a comprehensive health check"** - Overall system health - **"Show me node information with resource utilization"** - Node usage analysis - **"Check API server performance metrics"** - API server health - **"Show me OVN-K pod resource usage"** - Pod-level analysis - **"Analyze OVS performance metrics"** - OVS dataplane analysis - **"Analyze sync duration performance"** - Control plane sync metrics - **"Run unified performance analysis"** - Comprehensive system analysis ### Natural Language Examples - "What's the current status of my cluster?" - "Show me the top 5 pods using the most CPU" - "Are there any performance issues with the API server?" - "Generate a performance report for the last 2 hours" - "Which nodes are under the most stress?" - "How is the OVN synchronization performing?" ## Agent Capabilities The LangGraph agent provides: 1. **Intelligent Tool Selection**: Automatically chooses appropriate tools based on user queries 2. **Parameter Generation**: Creates correct parameters for tool calls based on context 3. **Data Interpretation**: Analyzes results and provides actionable insights 4. **Memory Persistence**: Remembers conversation context across interactions 5. **Error Handling**: Gracefully handles failures and provides helpful guidance ## Configuration ### LLM Configuration By default, the client uses OpenAI's GPT-3.5-turbo. To use a different LLM, modify the `initialize_agent()` function in `ovnk_benchmark_mcp_client_chat.py`: ```python # For Anthropic Claude from langchain_anthropic import ChatAnthropic llm = ChatAnthropic(model="claude-3-sonnet-20240229") # For Google Gemini from langchain_google_genai import ChatGoogleGenerativeAI llm = ChatGoogleGenerativeAI(model="gemini-pro") # For local models via Ollama from langchain_ollama import ChatOllama llm = ChatOllama(model="llama2") ``` ### Memory Configuration The agent uses LangGraph's MemorySaver for conversation persistence. Memory is stored in-process and will be lost when the client restarts. For persistent memory, implement a database-backed checkpointer. ## Troubleshooting ### Connection Issues 1. Verify MCP server is running on port 8000 2. Check that MCP client is running on port 8001 3. Ensure no firewall blocking local connections ### Tool Execution Errors 1. Verify OpenShift cluster accessibility 2. Check Prometheus connectivity 3. Validate authentication credentials ### Memory Issues 1. Conversation memory is per-session 2. Use unique conversation_id for different contexts 3. Clear memory by restarting the client ## Development ### Adding New Tools 1. Add tool functions to the MCP server 2. Create corresponding LangChain @tool wrappers in the client 3. Add tools to the agent's tool list in `initialize_agent()` ### Custom Analysis The agent can be extended with custom analysis capabilities by adding new tool functions that call the MCP server's analysis endpoints. ## API Endpoints ### MCP Client Endpoints - `GET /api/health` - Health status - `POST /chat/stream` - Streaming chat - `POST /api/tools/{tool_name}` - Direct tool calls ### MCP Server Endpoints (via tools) - Health and cluster information tools - Performance analysis tools - Historical data storage and retrieval - Comprehensive analysis and reporting ## License This project is part of the OpenShift OVN-K Benchmark suite and follows the same licensing terms.