MCP Server Implementation – Bug Triage Tracker

Overview

This Bug Triage Tracker implements a fully autonomous Model Context Protocol (MCP) server that enables multi-agent collaboration for bug classification and severity assessment. It now features a real-time analytics dashboard powered by MongoDB, offering full visibility into bug trends, team workload, and AI model performance.

MCP Architecture

Agent Roles

• Technical Analyst Agent (DeepSeek)

  • Analyzes technical aspects
  • Identifies affected components
  • Assesses technical severity
  • Determines required skills

• Context Understanding Agent (Llama)

  • Evaluates business impact
  • Assesses affected users and revenue
  • Determines business severity

• Decision Making Agent (Mistral)

  • Synthesizes inputs
  • Resolves conflicts
  • Assigns to the right developer

MCP Communication Protocol

class MessageType(Enum):
    ANALYZE = "analyze"
    QUERY = "query"
    RESPOND = "respond"
    VOTE = "vote"
    DECIDE = "decide"

Agent Collaboration Flow

1. Orchestrator → Technical Analyst: Request analysis
2. Analyst responds with category/severity
3. Orchestrator → Context Expert: Assess business impact
4. Context Expert responds
5. Analyst ↔ Context Expert: Discuss conflicting results
6. Orchestrator → Decision Agent: Make final call
7. Decision Agent → Orchestrator: Returns final triage decision

Key Features

1. Autonomous Severity Assessment

   • Agents assess independently
   • Business and technical impacts merged for consensus

2. Inter-Agent Communication

   • Structured messages
   • Disputes handled via dialogue

3. Decision Justification

   • Decision agent logs reasoning
   • Human-readable summaries

4. Real-Time Analytics with MongoDB

   • Tracks total bugs processed, model confidence, and processing time
   • Logs every decision and conflict resolution
   • Calculates ensemble accuracy dynamically
   • Monitors agent usage frequency
   • Visualizes team workload & utilization across categories

Implementation Details

MCP Message Format

@dataclass
class MCPMessage:
    sender: str
    receiver: str
    type: MessageType
    content: Dict
    context: Optional[Dict]

Knowledge Base Examples

• Technical Analyst

{
  "technical_category": "API",
  "complexity": "High",
  "components_affected": ["payment-service"],
  "severity_assessment": "High"
}

• Context Expert

{
  "user_impact": "Severe",
  "business_priority": "P0",
  "revenue_impact": "Direct - Critical",
  "severity_from_impact": "Critical"
}

MongoDB Integration (NEW)

• Collections: bug_collection, team_collection

• Stored Fields: created_at, processing_time, confidence, correct, models_used

• Backend Analytics:

• Accuracy = % of bugs marked correct

• Confidence = average across all predictions

• Real-time trend: daily bug count

• Model usage breakdown: DeepSeek, Llama, Mistral

• Team workload: current vs. max capacity

API Endpoints

/triage-bug-mcp

Main MCP endpoint Request:

{
  "title": "Payment gateway timeout",
  "description": "504 errors after 30 seconds",
  "severity": "High"
}

Response:

{
  "id": "BUG-20240605123456",
  "category": "API",
  "severity": "Critical",
  "priority": "P0",
  "assigned_to": {
    "name": "Alice Chen",
    "team": "API",
    "skill_match_score": 95
  },
  "mcp_details": {
    "messages_exchanged": 7,
    "collaboration_occurred": true,
    "communication_log": [...]
  }
}

/mcp-demo

Demo endpoint simulating full agent workflow.

Qualifying for Track 1 and 3

• #mcp-server-track: Implements the full MCP spec, agent messaging, and autonomous triage pipeline
• #agent-demo-track: Rich real-time analytics and interactive front-end showcase multi-agent intelligence in action

Running the Server

modal deploy src/modal_backend/mcp_app.py

Test:

curl -X POST https://your-app.modal.run/triage-bug-mcp \
     -H "Content-Type: application/json" \
     -d '{"title": "Bug title", "description": "Bug details"}'

Real-Time Dashboard

• Open Hugging Face Space to:

  • Trigger bugs
  • Monitor charts for severity & category trends
  • View model performance and system health
  • Watch Demo Video

Future Enhancements

• Agent memory from past bugs
• Auto-retraining for model drift
• Role-based access to analytics
• Integration with GitHub Issues and Jira