docs/module-5/Readme.md

Module 5: Building Agent Logic & MCP Integration

Overview

This module focuses on implementing the core logic of the Topcoder Challenge Steward Agent to communicate effectively with the Topcoder MCP server, retrieve challenge data, and generate intelligent, personalized outputs for users.

Project Structure

Core Architecture

tc-agent/
├── app.py                      # FastAPI main application
├── config/
│   └── settings.py            # Configuration management
├── services/
│   ├── mcp_service.py         # MCP client management
│   ├── challenge_service.py   # Challenge processing logic
│   ├── ai_service.py          # AI agent orchestration
│   ├── challenge_agent.py     # Agent service for notifications
│   └── scheduler.py           # Daily notification scheduler
├── prompts/
│   └── challenge_prompts.py   # Prompt engineering templates
├── tools/
│   └── email_tool.py          # Email notification tool
└── utils/
    ├── json_parser.py         # Response parsing utilities
    ├── error_handler.py       # Error handling
    └── logger.py              # Logging configuration

MCP Integration Implementation

1. Connection Method: Server-Sent Events (SSE)

We chose SSE over Streamable HTTP for real-time, persistent connections to the Topcoder MCP server.

Key Implementation (services/mcp_service.py):

from strands.tools.mcp import MCPClient
from mcp.client.sse import sse_client

class MCPService:
    def create_topcoder_client(self) -> MCPClient:
        """Create a Topcoder MCP client instance"""
        logger.info(f"Creating Topcoder MCP client for endpoint: {settings.TOPCODER_MCP_ENDPOINT}")
        return MCPClient(lambda: sse_client(settings.TOPCODER_MCP_ENDPOINT))

Configuration:

• Endpoint: https://api.topcoder-dev.com/v6/mcp/sse
• Transport: Server-Sent Events for real-time data streaming
• Authentication: No authentication required (public API)

2. MCP Tool Integration

The agent dynamically discovers and utilizes available MCP tools from the Topcoder server:

def get_topcoder_tools(self, client: MCPClient) -> List[Any]:
    """Get available Topcoder MCP tools from the client"""
    try:
        tools = client.list_tools_sync()
        tool_names = [getattr(tool, 'name', str(tool)) for tool in tools]
        logger.info(f"Available Topcoder MCP tools: {tool_names}")
        return tools
    except Exception as e:
        logger.error(f"Error getting Topcoder MCP tools: {e}")
        return []

3. Challenge Processing Logic

The core challenge processing is handled in services/challenge_service.py:

Dry Run Mode (Interactive Testing)

def get_challenges_dry_run(self, preferences: str) -> Dict:
    """Get challenges for dry run (immediate response without sending emails)"""
    try:
        # Create MCP client
        topcoder_client = self.mcp_service.create_topcoder_client()
        
        with topcoder_client:
            # Get available tools
            mcp_tools = self.mcp_service.get_topcoder_tools(topcoder_client)
            
            # Create AI agent with tools
            system_prompt = ChallengePrompts.get_dry_run_prompt(preferences)
            agent = self.ai_service.create_agent(mcp_tools, system_prompt)
            
            # Process user query
            response = agent(user_query)
            
            # Parse and normalize response
            parsed_data = self.json_parser.extract_json_from_response(str(response))
            challenges = self.json_parser.normalize_challenges(parsed_data["challenges"])
            
            return StandardAPIResponse.success(data={"challenges": challenges})
    except Exception as e:
        return handle_service_error(e, "get challenges for dry run", preferences)

Automated Processing Mode (Email Notifications)

def process_challenges_for_user(self, user_id: str, email: str, preferences: str) -> Dict:
    """Process challenges for automated agent (email notifications)"""
    try:
        topcoder_client = self.mcp_service.create_topcoder_client()
        
        with topcoder_client:
            mcp_tools = self.mcp_service.get_topcoder_tools(topcoder_client)
            
            # Combine MCP tools with email tool
            all_tools = [send_email] + mcp_tools
            
            # Create agent with personalized prompt
            system_prompt = ChallengePrompts.get_agent_prompt(preferences, email)
            agent = self.ai_service.create_agent(all_tools, system_prompt)
            
            # Execute automated workflow
            response = agent(user_query)

Prompt Engineering

1. Dry Run Prompt Template

For interactive testing and immediate responses:

def get_dry_run_prompt(preferences: str) -> str:
    return f"""
    You are a Topcoder Challenge Steward Agent specializing in finding relevant programming challenges.

    User Preferences: {preferences}

    Your tasks:
    1. Query active Topcoder challenges using available MCP tools
    2. Filter challenges based on user preferences (technologies, prize amounts, difficulty)
    3. Return results in structured JSON format

    Response Format:
    {{
        "challenges": [
            {{
                "id": "challenge_id",
                "title": "Challenge Title",
                "prize": "$X,XXX",
                "technologies": ["tech1", "tech2"],
                "registrationEndDate": "YYYY-MM-DD",
                "submissionEndDate": "YYYY-MM-DD",
                "url": "challenge_url",
                "description": "brief description"
            }}
        ]
    }}

    Focus on challenges that best match the user's stated preferences.
    """

2. Agent Prompt Template

For automated email notifications:

def get_agent_prompt(preferences: str, email: str) -> str:
    return f"""
    You are a professional Topcoder Challenge Steward Agent that helps developers find relevant programming challenges.

    User: {email}
    Preferences: {preferences}

    Your workflow:
    1. Query active Topcoder challenges using MCP tools
    2. Intelligently filter based on user preferences
    3. If relevant challenges found (2-5 high-quality matches), send a professional email
    4. Include challenge details, prize information, and registration deadlines

    Email Guidelines:
    - Professional, concise tone
    - Highlight why each challenge matches their preferences
    - Include prize amounts and key deadlines
    - Provide direct registration links
    - Only send if genuinely relevant challenges exist
    """

Data Processing & Response Generation

1. JSON Response Parsing

Robust parsing handles various MCP response formats:

class JSONParser:
    def extract_json_from_response(self, response: str) -> dict:
        """Extract JSON data from agent response, handling various formats"""
        # Remove markdown code blocks and whitespace
        # Parse JSON with error handling
        # Return structured data or None
        
    def normalize_challenges(self, challenges_data) -> List[Dict]:
        """Normalize challenge data to consistent format"""
        # Handle different challenge data structures
        # Ensure required fields are present
        # Apply consistent formatting

2. Error Handling & Resilience

Comprehensive error handling ensures reliable operation:

def handle_service_error(error: Exception, operation: str, context: str = None) -> Dict:
    """Standardized error handling for service operations"""
    return StandardAPIResponse.error(
        message=f"Failed to {operation}",
        error_code="SERVICE_ERROR",
        context={"operation": operation, "context": context}
    )

Performance Optimization for Hugging Face CPU Basic

1. Efficient Resource Usage

• Asynchronous Operations: All database operations use async/await
• Connection Pooling: MCP clients are created on-demand and properly closed
• Memory Management: Responses are processed and cleaned up promptly

2. CPU-Optimized Processing

# No GPU dependencies
# Lightweight JSON parsing
# Efficient string processing
# Minimal memory footprint for challenge data

3. Scalable Architecture

• Singleton MCP Service: Reuses client configurations
• Centralized Error Handling: Consistent error responses
• Modular Design: Easy to extend and maintain

Key Features Implemented

1. Real-time Challenge Discovery: Uses MCP SSE connection for live data
2. Intelligent Filtering: AI-powered preference matching
3. Dual Operation Modes: Interactive (dry-run) and automated (scheduled)
4. Robust Error Handling: Graceful degradation and informative error messages
5. Structured Responses: Consistent JSON format for frontend consumption
6. Email Integration: Automated notifications for relevant challenges
7. Performance Optimized: Designed for Hugging Face CPU Basic environment

Testing & Validation

Local Testing

# Start the application
python app.py

# Test dry-run endpoint
curl -X POST "http://localhost:8000/api/topcoder-dry-run" \
  -H "Content-Type: application/json" \
  -d '{"preferences": "Python, machine learning, $1000+ prize"}'

MCP Connection Verification

The agent automatically validates MCP connectivity on startup and logs available tools for debugging.

Next Steps

Module 5 successfully implements the core agent logic with robust MCP integration. The next module will focus on creating an intuitive user interface to interact with these backend services.