🚀 Initialize MCP - AI Development Assistant

The most advanced AI development memory and learning system.

📋 Quick Start

At the Start of Each Task

mcp_init({ task: "your task description" })

This displays the MCP Dashboard with:

• System health check
• Quick stats
• Mode selection (auto-selects after 5 seconds)

After Making Decisions

log_decision({
  type: "implementation",
  question: "What was decided?",
  chosen: "What was chosen",
  reasoning: "Why this choice",
  confidence: 0.9
})

🎯 MCP Init Dashboard

Features

Dashboard Display

┌────────────────────────────────────────────────────────────────────┐
│  🤖 GoGrab AI Development Assistant                                │
│  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
│                                                                    │
│  📊 System Status: ✅ Backend (42ms)    ✅ MCP v1.0.0             │
│  ┌─────────────┬─────────────┬─────────────┬─────────────┐        │
│  │ Decisions   │ Embeddings  │ Coverage    │ Golden      │        │
│  │ 9,962       │ 6,202       │ 62%         │ 200         │        │
│  └─────────────┴─────────────┴─────────────┴─────────────┘        │
│                                                                    │
│  🎯 Session: 10.7                                                  │
│                                                                    │
│  ▶ [1] 🚀 Quick Task (recommended)                                 │
│    [2] 🎭 Multi-Agent Orchestration                                │
│    [3] 🔍 Search Past Decisions                                    │
│    [4] 📈 View Learning Stats                                      │
│                                                                    │
│  💡 Recent Context:                                                 │
│  • Data architecture standardization                               │
│  • Multi-agent orchestration analysis                              │
│  • MCP init dashboard implementation                               │
│                                                                    │
│  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ │
│                                                                    │
│  💬 Reply 1, 2, 3, 4 — or say "go" for Quick Task                  │
│     Default: [1] 🚀 Quick Task                                     │
│                                                                    │
└────────────────────────────────────────────────────────────────────┘

Mode Selection

Smart UX

• 5-second auto-select: Proceeds with default if no response
• Double ENTER: Quick confirm for default selection
• Task complexity detection: Analyzes task → suggests appropriate mode

🔧 Available MCP Tools

Core Tools

Advanced Tools

📊 3-Tier Data Architecture

Data Flow

┌─────────────────────────────────────────────────────────────────┐
│                     TIER 1: RAW DECISIONS                        │
│                     ai-agents.decisions                          │
│                     (9,962 documents)                            │
└─────────────────────────┬───────────────────────────────────────┘
                          │ Auto-embed
                          ▼
┌─────────────────────────────────────────────────────────────────┐
│                  TIER 2: VECTOR EMBEDDINGS                       │
│                  ai-memory.agent_embeddings                      │
│                  (6,202 documents - 62% coverage)                │
└─────────────────────────┬───────────────────────────────────────┘
                          │ Promote (high confidence, outcomes)
                          ▼
┌─────────────────────────────────────────────────────────────────┐
│                    TIER 3: GOLDEN DECISIONS                      │
│                    ai-agents.golden_decisions                    │
│                    ai-memory.golden_embeddings                   │
│                    (200 documents each)                          │
└─────────────────────────────────────────────────────────────────┘

Databases

Cross-Reference Field

All collections use originalDecisionId for linking:

// Find embedding for a decision
decisions._id = "abc123"
agent_embeddings.originalDecisionId = "abc123"
golden_decisions.originalDecisionId = "abc123"
golden_embeddings.originalDecisionId = "abc123"

🧠 Intelligence Pyramid

Context Loading Priority

       ┌─────────────┐
       │ CORE RULES  │  ← Always loaded (~20 items, ~500 tokens)
       │  Priority 1 │
       └──────┬──────┘
              │
       ┌──────▼──────┐
       │  PATTERNS   │  ← Loaded for matching categories
       │  Priority 2 │
       └──────┬──────┘
              │
       ┌──────▼──────┐
       │  SIMILAR    │  ← Vector search for specific query
       │  Priority 3 │
       └──────┬──────┘
              │
       ┌──────▼──────┐
       │  WORKING    │  ← Current session memory
       │   MEMORY    │
       └─────────────┘

Token Budget

🎭 Multi-Agent Orchestration

Agent Types

Orchestration Flow

Task: "Implement user authentication"
                │
                ▼
┌─────────────────────────────────────────────────────────────┐
│  1. DECOMPOSITION                                            │
│     Task → [Setup Auth Module] → [Implement Login] →        │
│            [Add Guards] → [Test & Verify]                   │
└─────────────────────────┬───────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────────┐
│  2. AGENT SELECTION                                          │
│     Step 1: initializer (setup)                              │
│     Step 2: specialist/security (login)                      │
│     Step 3: executor (guards)                                │
│     Step 4: reviewer (testing)                               │
└─────────────────────────┬───────────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────────┐
│  3. EXECUTION                                                │
│     Each step creates session → decisions → outcomes        │
└─────────────────────────────────────────────────────────────┘

API Endpoints

🔄 Adaptive Learning Triggers

Trigger Types

Learning Jobs

📁 File Structure

MCP Source

MCP/mcp-development-ai/src/
├── server.ts                    # Main server & tool registration
├── config/
│   └── constants.ts             # Configuration values
├── storage/
│   ├── file-storage.ts          # File persistence
│   ├── memory-store.ts          # In-memory store
│   └── docs-data-integration.ts # Session data integration
├── tools/
│   ├── mcp-init.ts              # 🆕 Init dashboard
│   ├── preflight.ts             # Context loading
│   ├── log-decision.ts          # Decision logging
│   ├── find-similar.ts          # Semantic search
│   └── ...                      # Other tools
└── services/
    ├── decision-lifecycle-manager.ts
    └── outcome-validator.ts

Session Data

sessions/
├── sessionID.json               # Current session tracking
├── sessions-history/
│   ├── session-9.json
│   └── session-10.json
└── schemas/
    ├── session-manifest.schema.json
    └── session-manifest.template.json

Local Cache

.cursor/ai-memory/
├── current-session.json         # Hot session data
├── working-memory.json          # Quick summaries
├── conventions.json             # Learned conventions
└── sessions/
    └── session-9.json           # Historical cache

🔗 Backend API Endpoints

Infrastructure

Embeddings

Migration/Cleanup

📅 Date Standardization

Standard Fields

All collections use consistent date fields:

Shared Utility

import { formatDisplayDate, createTimestampFields } from '@app/shared/date.utils';

// Format for display
formatDisplayDate(new Date()) // "Dec 17, 2025, 10:51 AM"

// Create all timestamp fields
createTimestampFields() // { timestamp, displayDate, createdAt, updatedAt }

⚡ Quick Reference Commands

Terminal Commands

# Check backend health
curl -s "http://localhost:3001/ai-api/infrastructure/health" | jq '.'

# Get pipeline stats
curl -s "http://localhost:3001/ai-api/infrastructure/pipeline/stats" | jq '.'

# Check session
cat sessions/sessionID.json

# Session manager status
node tools/scripts/database/session-manager.js status

MCP Tool Calls

// Initialize session
mcp_init({ task: "Fix authentication bug" })

// Get context before decision
preflight({ task: "How to structure auth module" })

// Log decision
log_decision({
  type: "architecture",
  question: "How to structure auth?",
  chosen: "Separate service + guards",
  reasoning: "Better testability and separation of concerns",
  confidence: 0.9,
  files: ["auth.service.ts", "auth.guard.ts"]
})

// Report outcome
report_outcome({
  decisionId: "dev_1765995218122_xhpxjtq",
  wasSuccessful: true,
  rating: 5,
  feedback: "Works perfectly"
})

🚨 Troubleshooting

Common Issues

Health Check

# Full system check
curl -s "http://localhost:3001/ai-api/infrastructure/health" | jq '.'

# Expected output:
{
  "status": "healthy",
  "databases": {
    "ai-agents": { "status": "healthy" },
    "ai-memory": { "status": "healthy" },
    "ai-mcp-dev": { "status": "healthy" }
  },
  "stats": {
    "decisions": 9962,
    "embeddings": 6202,
    "sessions": 0
  }
}

📈 System Statistics (Current)

🔮 Future Enhancements

• Real-time WebSocket updates for orchestration progress
• Visual execution graph for multi-agent tasks
• A/B testing for decision strategies
• Cross-project learning transfer
• Natural language task decomposition

Last Updated: December 17, 2025 - Session 10.7