Dewy Resort

Dewy Resort Hotel - Sample Application

Enterprise MCP Design Patterns for LLM Productivity

A demonstration of how to design Model Context Protocol (MCP) servers that maximize AI agent productivity while maintaining backend system integrity, performance, security, and scale.

---

What This Demonstrates

This sample application showcases enterprise-grade MCP architecture patterns that solve real-world challenges when integrating LLMs with backend systems:

The Problem

Most MCP implementations give AI agents direct database access or expose raw APIs, leading to:

• ❌ Poor performance - Agents make 10+ API calls for simple tasks
• ❌ Data integrity issues - Agents create invalid state transitions
• ❌ Security vulnerabilities - Overly permissive tool access
• ❌ Scalability bottlenecks - No caching, retries, or rate limiting
• ❌ Poor user experience - Slow responses, frequent failures

The Solution: Compositional MCP Architecture

This project demonstrates a two-tier tool design that balances LLM flexibility with enterprise requirements:

🎯 Orchestrators (High-level tools) - Optimized for common scenarios

• Encode business rules and prerequisites
• Execute multi-step workflows in

Enterprise MCP Architecture - Persona-based servers with orchestrators and atomic skills

Real-World Example: Guest Check-In

*Check-in orchestrator: ✅ Atomic skills for flexibility - Handle edge cases intelligently ✅ Zero direct integrations - All through central hub ✅ Persona-based servers - Security, performance, usability ✅ Idempotency and validation - Data integrity, reliability ✅ Performance optimization - Parallel reads, dependency ordering

---

Additional Resources

Setup Guides

• [Salesforce Setup](./vendor/salesforce/docs/SALESFORCE_SETUP.md) - Deploy custom objects and seed data
• [Workato Setup](./vendor/workato/docs/WORKATO_SETUP.md) - Deploy MCP server recipes
• [Architecture Diagrams](./app/docs/architecture/README.md) - Visual documentation of all workflows

Optional Features

• [Stripe Integration](./vendor/workato/docs/WORKATO_SETUP.md#stripe-recipe-activation-optional) - Payment processing
• [Cognito Authentication](#cognito-authentication-workshop-convenience) - User auth (workshop convenience)
• [Bedrock AI Chat](#bedrock-ai-chat-optional) - AI assistants (optional)

Technical Details

• [Salesforce Metadata](./vendor/salesforce/README.md) - Complete object and field documentation
• [Project Structure](#project-structure) - Codebase organization
• [CLI Commands](#cli-commands) - Automation scripts

---

Project Structure

dewy-resort/
├── workato/                      # ⭐ MCP SERVER IMPLEMENTATION
│   ├── recipes/
│   │   ├── atomic-salesforce-recipes/  # 15 atomic skills
│   │   ├── atomic-stripe-recipes/      # 6 payment atomic skills
│   │   ├── orchestrator-recipes/       # 13 high-level orchestrators
│   │   ├── home-assistant/             # Home Assistant integration
│   │   ├── sf-api-collection/          # API Collection definitions
│   │   └── Workspace Connections/      # Connection configs
├── app/                          # Next.js application
│   ├── src/                      # Application source
│   ├── docs/
│   │   └── architecture/         # ⭐ ARCHITECTURE DIAGRAMS
│   │       ├── system-architecture.png
│   │       ├── guest-checkin-flow.png
│   │       ├── guest-checkout-flow.png
│   │       ├── guest-service-request-flow.png
│   │       └── maintenance-request-flow.png
│   └── public/                   # Static assets
├── vendor/
│   ├── workato/                  # Workato docs & scripts
│   │   ├── docs/
│   │   │   └── WORKATO_SETUP.md
│   │   └── scripts/
│   ├── salesforce/               # Salesforce metadata & deployment
│   │   ├── force-app/            # Custom objects, fields, app
│   │   ├── data/                 # Seed data
│   │   ├── docs/
│   │   │   └── SALESFORCE_SETUP.md
│   │   └── scripts/
│   │       └── deploy.sh
│   ├── aws/                      # CloudFormation templates
│   └── okta/                     # Okta auth config

---

CLI Commands

# Setup
make setup                        # Verify CLI installations
make workato-login                # Authenticate wk CLI with API token

# Workato (MCP Server) — requires wk CLI (brew install workato-devs/tap/wk)
make workato-init                 # Initialize wk project & pull recipes
make validate                     # Lint all recipes
make push                         # Push recipes to workspace
make pull                         # Pull recipes from workspace
make start-recipes                # Start all recipes
make stop-recipes                 # Stop all recipes
make setup-api                    # Create API collections, endpoints & client
make enable-api-endpoints         # Enable API endpoints
make setup-mcp                    # Create MCP servers & write config to app/.env

# Salesforce (Backend)
make sf-deploy org=        # Deploy metadata and seed data

# Diagnostics
make status                       # Check all CLI status
make doctor                       # Verify CLI installations

---

Mock Mode (Development Only)

For frontend development without backend setup:

# In .env
WORKATO_MOCK_MODE=true

# Restart server
npm run dev

⚠️ Note: Mock mode simulates MCP responses. Use for frontend work only, not for learning MCP architecture patterns.

---

Optional Features

Cognito Authentication (Workshop Convenience)

For workshops where participants don't have their own auth:

AUTH_PROVIDER=cognito
# Deploy Cognito User Pool
cd aws/cloudformation
./deploy.sh dev http://localhost:3000/api/auth/cognito/callback http://localhost:3000 dewy-hotel

Bedrock AI Chat (Optional)

AI-powered chat assistants (demonstrates LLM + MCP integration):

# Deploy Identity Pool
cd aws/cloudformation
./deploy-identity-pool.sh dev  

# Configure in .env
COGNITO_IDENTITY_POOL_ID=your_pool_id

---

Technology Stack

• MCP Server: Workato (33 recipes organized as orchestrators and atomic skills)
• Backend Systems: Salesforce (CRM), Stripe (payments), Twilio (SMS)
• Application: Next.js 14, React, TypeScript, Tailwind CSS
• Database: SQLite (local app data only)
• Auth: Amazon Cognito (optional) or mock mode
• AI: Amazon Bedrock (optional chat assistants)

---

Why Workato?

This sample uses Workato as the MCP server implementation, but the architectural patterns apply to any integration platform:

✅ Visual recipe builder - Easy to understand workflows ✅ Built-in connectors - Salesforce, Stripe, Twilio out of the box ✅ API Collections - Native REST API exposure ✅ Enterprise features - Error handling, retries, logging, monitoring ✅ Workshop-friendly - Visual representation aids learning

The patterns work with: Custom APIs, serverless functions, other orchestration systems, etc.

---

License

MIT

---

Questions?

This is a sample application for teaching enterprise MCP design patterns. The goal is to demonstrate how to build MCP servers that maximize LLM productivity while maintaining backend integrity, performance, security, and scale.

Focus areas:

• Why orchestrators matter for performance and UX
• When to use atomic skills vs orchestrators
• How to design persona-based MCP servers
• Patterns for idempotency and state validation
• Zero direct system integrations architecture

For implementation details, see the setup guides in [vendor/workato/docs/](./vendor/workato/docs/) and [vendor/salesforce/docs/](./vendor/salesforce/docs/).

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: workato-devs
• Source: workato-devs/dewy-resort
• License: MIT

Install and usage instructions live in the source repository linked above.