MCP Development AI - Complete Setup & Verification

This guide helps you set up and verify the Tier 1: Development AI MCP server that enables Claude in Cursor to learn from your coding sessions.

Current Status

✅ Built: MCP server compiled successfully (dist/server.js exists)
❌ Not Configured: Not added to Cursor's MCP configuration yet
❌ Not Running: No active connection yet

Quick Setup (5 minutes)

Step 1: Update Cursor MCP Configuration

File: .cursor/mcp.json

Current state:

{
  "mcpServers": {}
}

Update to:

{
  "mcpServers": {
    "mcp-development-ai": {
      "command": "node",
      "args": [
        "/Users/fguijarro/developer/repos/editor2/MCP/mcp-development-ai/dist/server.js"
      ],
      "env": {
        "MONGODB_CONNECTION_STRING": "your-mongodb-connection-string"
      }
    }
  }
}

Important: Use your actual MongoDB connection string from .env

Step 2: Restart Cursor

1. Close all Cursor windows
2. Restart Cursor
3. Open this workspace again

Step 3: Verify MCP Connection

Open a new chat with Claude and ask:

Can you call the get_stats MCP tool to show me Development AI stats?

Expected response:

{
  "totalDecisions": 0,
  "successRate": 0,
  "conventionCount": 0,
  "patternCount": 0,
  "recentDecisions": 0
}

If Claude can respond with this, the MCP server is working! 🎉

What This MCP Server Does

🎯 Purpose: Tier 1 Learning

Makes Claude remember what works in THIS codebase specifically.

🔧 Available Tools

📂 Storage Locations

Local (Primary):

• Location: .cursor/ai-memory/development-memory.json
• Created on first use
• Works offline
• Fast access

MongoDB (Optional Backup):

• Database: ai-mcp-dev
• Collections: decisions, conventions, patterns
• Syncs automatically if connection string provided
• Enables cross-machine learning

Testing the MCP Server

Test 1: Check Stats (Basic Connection)

Ask Claude:

Use the get_stats MCP tool

Expected: Should return JSON with current stats (all zeros initially)

Test 2: Log a Decision

Ask Claude:

Use the log_decision MCP tool to record this test decision:
- Type: "implementation"
- Question: "Test MCP logging"
- Options: ["Option A", "Option B"]
- Chosen: "Option B"
- Reasoning: "Testing the MCP server setup"
- Confidence: 0.9

Expected:

{
  "id": "dev_1733887200000_abc123",
  "success": true
}

Test 3: Retrieve the Decision

Ask Claude:

Use find_similar MCP tool to search for "test MCP"

Expected: Should return the decision you just logged

Test 4: Verify Storage File

# Check local storage
cat .cursor/ai-memory/development-memory.json

# Should show:
# {
#   "decisions": [{ ... your test decision ... }],
#   "conventions": [],
#   "patterns": [],
#   "metadata": { ... }
# }

How Claude Should Use This

✅ Good Usage Examples

Before Making Decisions:

Claude: Let me check if we've solved this before
[Calls find_similar tool]
Claude: Found a similar decision from last week...

After Making Decisions:

Claude: I'm choosing approach B because...
[Calls log_decision to record reasoning]

When Discovering Patterns:

Claude: I notice this codebase uses inject() instead of constructors
[Calls add_convention to store pattern]

❌ What NOT to Log

Don't log trivial decisions:

• ❌ "Fixed typo"
• ❌ "Added import statement"
• ❌ "Renamed variable from x to y"

DO log significant decisions:

• ✅ "Chose RxJS over Promises for async flow because..."
• ✅ "Used Signal instead of BehaviorSubject because..."
• ✅ "Implemented caching at service layer vs component level because..."

Troubleshooting

"MCP tool not available"

Problem: Cursor can't find the MCP server

Fix:

1. Check .cursor/mcp.json has the correct path
2. Verify dist/server.js exists: ls tools/mcp-development-ai/dist/
3. Restart Cursor completely (all windows)

"Error executing MCP tool"

Problem: Server crashes when calling tools

Fix:

1. Check Cursor's Developer Tools (Help → Toggle Developer Tools)
2. Look for MCP server errors in console
3. Verify node version: node --version (should be >= 18)

"MongoDB connection failed"

Problem: Optional MongoDB sync not working

Fix:

// This is OK! MCP works without MongoDB
// Just uses local storage instead

// If you want MongoDB sync:
1. Verify MONGODB_CONNECTION_STRING in .cursor/mcp.json
2. Test connection: node -e "const {MongoClient} = require('mongodb'); new MongoClient('your-uri').connect().then(() => console.log('OK'))"

"No decisions showing up"

Problem: Storage file empty

Check:

# File should exist after first log_decision call
ls -la .cursor/ai-memory/

# Read contents
cat .cursor/ai-memory/development-memory.json

Fix: Ask Claude to use log_decision tool manually (see Test 2)

Verification Checklist

• MCP server built (dist/server.js exists)
• Configuration added to .cursor/mcp.json
• Cursor restarted
• get_stats tool works (returns JSON)
• log_decision tool works (returns decision ID)
• find_similar tool works (finds logged decision)
• Storage file created (.cursor/ai-memory/development-memory.json)
• File contains logged decision

Integration with Full AI Learning System

┌─────────────────────────────────────────┐
│ Tier 1: Development AI (THIS MCP)       │  ← You are here
│ - Learns from Claude in Cursor          │
│ - Stores locally in .cursor/ai-memory/  │
│ - Optional MongoDB sync                 │
├─────────────────────────────────────────┤
│ Tier 2: Platform AI                     │
│ - Powers GoGrab Space AI features       │
│ - Uses backend API + vector search      │
│ - Stores in ai-agents database      │
├─────────────────────────────────────────┤
│ Tier 3: Tenant AI                       │
│ - Per-user personalization              │
│ - Tenant-isolated learning              │
└─────────────────────────────────────────┘

Tier 1 is independent - works even if backend is off!

Next Steps After Setup

1. Seed Initial Conventions

Ask Claude to analyze CLAUDE.md and AGENTS.md:

Read CLAUDE.md and use the add_convention MCP tool to store the main Angular/TypeScript rules we follow.

2. Start Logging Naturally

Claude should automatically use these tools during normal work:

• find_similar before making decisions
• log_decision after significant choices
• add_convention when discovering patterns

3. Review Learning Progress

Periodically ask:

Use get_stats to show how much Development AI has learned

4. Enable MongoDB Sync (Optional)

If you want cross-machine learning:

1. Add MONGODB_CONNECTION_STRING to mcp.json env
2. Decisions sync to ai-mcp-dev database automatically
3. Share learnings across your dev machines

Command Reference

# Build MCP server
cd MCP/mcp-development-ai
npm install
npm run build

# Watch mode (for development)
npm run dev

# Check build output
ls -la dist/

# View stored decisions
cat .cursor/ai-memory/development-memory.json | jq .

# Test MongoDB connection
node -e "const {MongoClient} = require('mongodb'); new MongoClient('your-connection-string').connect().then(() => console.log('✅ Connected')).catch(e => console.error('❌ Failed:', e.message))"

Created: December 10, 2025
Session: 13
Status: MCP Server Built ✅ | Configuration Needed ⚠️

Next Action: Update .cursor/mcp.json with the configuration above and restart Cursor!