krishvenkatraman-okta/Enterprise-AI-Platform-with-MCP-Server

Enterprise AI Platform with MCP Server

A comprehensive demo platform featuring dual interfaces: Atlas Beverages (inventory management) and J.A.R.V.I.S (AI assistant) with enterprise-grade Multi-Control Point (MCP) server architecture.

🚀 Quick Start (Zero Dependencies)

# Clone and run immediately - no database setup needed!
git clone https://github.com/yourusername/enterprise-ai-platform.git
cd enterprise-ai-platform
npm install
npm run dev

🚀 One-Click Deploy:

Local Development: npm run dev → http://localhost:5000
Production Deploy: Vercel/Netlify → https://your-app.vercel.app

Features included:

✅ Dual interface system - Atlas Beverages + J.A.R.V.I.S
✅ Working OAuth2 authentication - Complete login/logout flow
✅ Cross-application token exchange - JAG tokens for inventory access
✅ External LLM integration - Production-ready MCP endpoints
✅ Vercel deployment ready - Zero-config serverless functions
✅ Pre-populated demo data - 3 warehouses, 13 inventory items
✅ Real-time inventory queries - State-based warehouse filtering

🏗️ Architecture Overview

Latest Improvements (August 2025) 🆕

• ✅ Fixed Vercel deployment - Consolidated all endpoints into single api/server.js
• ✅ Resolved JSON parsing errors - Proper serverless function routing
• ✅ Fixed data structure mismatches - Frontend and backend now fully compatible
• ✅ Working logout functionality - Complete session management
• ✅ Enhanced external LLM integration - Production-ready OAuth2 endpoints
• ✅ Warehouse-specific queries - Precise state-based filtering
• ✅ Cross-interface authentication - Seamless JAG token exchange

Dual Interface System

• Atlas Beverages: Inventory management interface
• J.A.R.V.I.S: AI assistant with inventory access via JAG tokens

In-Memory Storage (Demo Ready)

• No database dependencies
• Realistic demo data pre-loaded
• Session-based persistence
• Perfect for GitHub/Vercel deployment

MCP Authorization Server

• OAuth 2.0 JWT-bearer grant type
• JAG token validation against Okta
• Cross-application token exchange
• External LLM integration support

📊 Demo Data

The platform starts with realistic inventory across 3 warehouses:

California Warehouse (Los Angeles)

• Premium Cola Classic (150 units)
• Craft IPA Selection (45 units)
• Energy Boost Original (200 units)
• Sparkling Water Lemon (30 units - LOW STOCK)
• Organic Green Tea (80 units)

Texas Warehouse (Austin)

• Sweet Tea Southern Style (120 units)
• Local Root Beer (85 units)
• Sports Hydration Blue (180 units)
• BBQ Cola Limited (25 units - LOW STOCK)

Nevada Warehouse (Las Vegas)

• Desert Spring Water (300 units)
• Premium Mixer Tonic (75 units)
• Luxury Vodka Selection (40 units)
• Casino Energy Rush (15 units - LOW STOCK)

🔐 Authentication Flows

Frontend (J.A.R.V.I.S & Atlas)

1. User authenticates via Okta OIDC
2. Receives JAG (JSON Authorization Grant) token
3. JAG token exchanged for MCP access token
4. Access inventory via MCP server

External LLM Integration 🤖

Production URLs (Vercel deployment):

# Step 1: Exchange JAG for MCP access token
curl -X POST https://your-app.vercel.app/api/oauth2/token \
  -H "Authorization: Basic $(echo -n 'mcp_inventory_server_001:mcp_server_secret_2024_inventory_access' | base64)" \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=urn:ietf:params:oauth:grant-type:jwt-bearer" \
  -d "assertion=<JAG_TOKEN>"

# Step 2: Query inventory with access token  
curl -X POST https://your-app.vercel.app/api/mcp/inventory/query \
  -H "Authorization: Bearer <ACCESS_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"type": "warehouse", "filters": {"state": "California"}}'

Local development:

# Same endpoints work locally on http://localhost:5000
curl -X POST http://localhost:5000/api/oauth2/token \
  -H "Authorization: Basic $(echo -n 'mcp_inventory_server_001:mcp_server_secret_2024_inventory_access' | base64)" \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=urn:ietf:params:oauth:grant-type:jwt-bearer" \
  -d "assertion=<JAG_TOKEN>"

🛠️ Tech Stack

Frontend

• React 18 + TypeScript
• Radix UI + shadcn/ui + Tailwind CSS
• TanStack Query for state management
• Wouter for routing

Backend

• Express.js + TypeScript
• In-memory storage (no database needed)
• Okta OAuth 2.0 integration
• MCP server with JWT validation

Development

• Vite for fast builds
• ESLint + TypeScript for code quality
• Hot reload for development

🌐 API Endpoints

Authentication & Sessions

• GET /api/config - Get Okta configuration (client IDs, domains)
• POST /api/auth/callback - OAuth callback handler with PKCE
• POST /api/auth/login - Process ID token and create session
• POST /api/auth/token-exchange - JAG to access token exchange (J.A.R.V.I.S)
• POST /api/auth/logout - Session termination and cleanup

Inventory Management

• GET /api/warehouses - List all warehouses
• GET /api/inventory/:warehouseId - Get warehouse inventory

MCP Server (External LLM Integration)

• POST /api/oauth2/token - OAuth2 JWT-bearer token exchange
• POST /api/mcp/inventory/query - Query inventory with access token
• GET /api/health - Server status and configuration

Note: All endpoints are consolidated into /api/server.js for Vercel serverless compatibility.

🔧 Environment Setup

Create .env file for production use:

# Okta Configuration
OKTA_DOMAIN=your-okta-domain.okta.com
OKTA_AUTHORIZATION_SERVER=https://your-okta-domain.okta.com/.well-known/oauth-authorization-server

# Client Credentials (or use demo defaults)
INVENTORY_CLIENT_ID=your_inventory_client_id
INVENTORY_CLIENT_SECRET=your_inventory_client_secret
JARVIS_CLIENT_ID=your_jarvis_client_id  
JARVIS_CLIENT_SECRET=your_jarvis_client_secret

# MCP Server Credentials (or use demo defaults)
MCP_SERVER_CLIENT_ID=mcp_inventory_server_001
MCP_SERVER_CLIENT_SECRET=mcp_server_secret_2024_inventory_access

Note: Demo works without environment setup using fallback credentials.

📋 Deployment

Vercel (Recommended) ⚡

Architecture: All API routes consolidated into single api/server.js file for serverless compatibility.

# Option 1: Deploy directly to Vercel
npm install -g vercel
vercel

# Option 2: Connect your GitHub repo to Vercel dashboard
# - Import your GitHub repository
# - Vercel automatically detects the configuration
# - Set environment variables in Vercel dashboard (optional)

File Structure (Vercel-optimized):

api/
├── server.js          # ← All endpoints (OAuth2 + MCP + Auth)
└── (no other files)   # ← Removed individual function files

vercel.json            # ← Routes all /api/* to server.js

Docker

# Build and run with Docker
docker build -t enterprise-ai-platform .
docker run -p 5000:5000 enterprise-ai-platform

Manual Deployment

# Build for production
npm run build

# Start production server
npm start

🧪 Testing the Integration

Test MCP Health 🩺

Production (Vercel):

curl https://your-app.vercel.app/api/health | jq .

Local Development:

curl http://localhost:5000/api/health | jq .

Test OAuth2 Token Exchange 🔐

Production (Vercel):

# Step 1: Get MCP access token using JAG (Production)
curl -X POST https://your-app.vercel.app/api/oauth2/token \
  -H "Authorization: Basic $(echo -n 'mcp_inventory_server_001:mcp_server_secret_2024_inventory_access' | base64)" \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=urn:ietf:params:oauth:grant-type:jwt-bearer" \
  -d "assertion=<REAL_JAG_TOKEN_FROM_JARVIS>"

Local Development:

# Step 1: Get MCP access token using JAG (Local)
curl -X POST http://localhost:5000/api/oauth2/token \
  -H "Authorization: Basic $(echo -n 'mcp_inventory_server_001:mcp_server_secret_2024_inventory_access' | base64)" \
  -H "Content-Type: application/x-www-form-urlencoded" \
  -d "grant_type=urn:ietf:params:oauth:grant-type:jwt-bearer" \
  -d "assertion=<REAL_JAG_TOKEN_FROM_JARVIS>"

Expected response (both environments):

{
  "token_type": "Bearer",
  "access_token": "abc123def456...",
  "expires_in": 86400,
  "scope": "inventory:read"
}

Test Inventory Query 📊

Production (Vercel):

# Step 2: Query California warehouse (Production)
curl -X POST https://your-app.vercel.app/api/mcp/inventory/query \
  -H "Authorization: Bearer <ACCESS_TOKEN_FROM_STEP_1>" \
  -H "Content-Type: application/json" \
  -d '{"type": "warehouse", "filters": {"state": "California"}}'

Local Development:

# Step 2: Query California warehouse (Local)
curl -X POST http://localhost:5000/api/mcp/inventory/query \
  -H "Authorization: Bearer <ACCESS_TOKEN_FROM_STEP_1>" \
  -H "Content-Type: application/json" \
  -d '{"type": "warehouse", "filters": {"state": "California"}}'

Expected response (both environments):

{
  "success": true,
  "data": {
    "warehouse": {"id": "warehouse-ca-001", "name": "West Coast Distribution"},
    "items": [...],
    "totalItems": 5,
    "lowStockItems": [...]
  }
}

📚 Documentation

• - External LLM integration guide
• - Complete MCP API reference
• - Quick testing guide

🤝 Contributing

This is a demo application showcasing enterprise-grade patterns:

1. Fork the repository
2. Create your feature branch
3. Make your changes (data persists during session)
4. Test with the MCP endpoints
5. Submit a pull request

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

Perfect for demos, POCs, and showcasing enterprise AI integration patterns without complex database setup.