DineAgent

AI-powered restaurant website platform. Restaurant websites aren't brochures — they're AI employees that understand menus, manage reservations, and handle reviews.

What is DineAgent?

DineAgent is a full-stack platform that turns a restaurant website into an intelligent, manageable system. Every business operation (menu, reservations, orders, events, reviews, loyalty, SEO) is exposed through three interfaces:

Interface	Use Case
MCP Server	AI agents (Claude, Cognix) call tools directly
CLI	Humans, scripts, and Cognix skill handlers
Payload CMS Admin	Visual web dashboard for non-technical staff

All three interfaces share the same data layer — changes from any source are immediately visible everywhere.

Architecture

dineagent/
├── apps/
│   ├── web/           # Astro frontend (restaurant website)
│   ├── cms/           # Payload CMS (admin + API)
│   ├── mcp/           # MCP Server (AI agent interface)
│   └── cli/           # CLI (human + script interface)
├── packages/
│   ├── shared/        # Common types and utilities
│   └── components/    # Business logic (core architecture)
│       ├── menu/           # Menu management
│       ├── reservation/    # Reservation management
│       ├── review/         # Review management
│       ├── order/          # Online ordering
│       ├── event/          # Events & banquets
│       ├── loyalty/        # Loyalty program
│       └── seo/            # SEO management
├── skills/            # Auto-generated Cognix skill packages
└── tooling/           # Build scripts & generators

Component Architecture

Each business component follows a strict 6-file contract:

menu/
├── component.yaml     # Metadata, MCP tools, CLI commands, Cognix bridge config
├── schema.ts          # Payload CMS CollectionConfig
├── mcp-tools.ts       # Zod schemas + MCP tool handlers
├── cli-commands.ts    # Commander.js command definitions
├── types.ts           # TypeScript domain types
└── index.ts           # Barrel exports

Key principle: MCP tools and CLI commands are thin adapters over the same Payload CMS operations. No business logic lives in the interface layer.

HITL (Human-in-the-Loop) Workflow

AI/CLI creates → status: draft → Admin reviews → status: published → Website renders
                                     ↓
                              status: archived (rejected)

Every collection has status (draft/published/archived) and createdBy (mcp-agent/cli/human) fields.

Quick Start

Prerequisites

• Node.js >= 18
• pnpm >= 9

Install

git clone https://github.com/codeducker/dineagent.git
cd dineagent
cp .env.example .env
pnpm install

Configure

Edit .env:

PAYLOAD_SECRET=your-random-secret-here
DATABASE_URL=./dineagent.db
PORT=3000
PAYLOAD_API_URL=http://localhost:3000/api

Build

pnpm build

Run

# Start CMS (admin panel + API)
cd apps/cms && pnpm dev
# Admin: http://localhost:3000/admin

# Start Astro frontend (in another terminal)
cd apps/web && pnpm dev
# Website: http://localhost:4321

Usage

CLI Commands

# Menu
dineagent menu create-dish --name "Spring Lamb" --price 28 --description "Herb-crusted"
dineagent menu list-dishes --format table
dineagent menu update-dish --id <id> --price 32

# Reservations
dineagent reservation create --guest-name "John" --party-size 4 --date 2026-05-15 --time 19:00
dineagent reservation list --date 2026-05-15
dineagent reservation cancel --id <id>

# Orders
dineagent order create --customer-name "Jane" --items '[{"dishName":"Pasta","quantity":2,"unitPrice":18}]' --order-type takeout
dineagent order list --status pending

# Events
dineagent event create --name "Wine Tasting" --date 2026-06-01 --start-time 18:00 --event-type private_dining
dineagent event list --format table

# Reviews
dineagent customer-review create --guest-name "Alice" --rating 5 --content "Amazing food!"
dineagent customer-review list --min-rating 4

# Loyalty
dineagent loyalty create-member --name "Bob" --email "bob@example.com"
dineagent loyalty add-points --id <id> --points 500 --description "Dinner visit"
dineagent loyalty list-members --format table

# SEO
dineagent seo get-meta --path /menu
dineagent seo update-meta --path /menu --title "Our Menu" --description "Fresh seasonal dishes"
dineagent seo update-structured-data --name "DineAgent Restaurant" --cuisine "Contemporary"

# HITL Review
dineagent review list
dineagent review approve <id> --collection dishes
dineagent review reject <id> --collection dishes

# Site Context
dineagent site context
dineagent site update-context --brand-voice "Warm, welcoming, seasonal"

MCP Server (AI Agents)

Configure in Claude Desktop or any MCP client:

{
  "mcpServers": {
    "dineagent": {
      "command": "node",
      "args": ["apps/mcp/dist/index.js"]
    }
  }
}

Available MCP tools:

Tool	Description
menu.create_dish	Create a new dish (as draft)
menu.list_dishes	List dishes with filters
menu.update_dish	Update a dish
menu.delete_dish	Delete a dish
menu.list_categories	List menu categories
reservation.create	Create a reservation
reservation.list	List reservations
reservation.update	Update a reservation
reservation.cancel	Cancel a reservation
reservation.availability	Check table availability
review.create	Create a review
review.list	List reviews
review.respond	Respond to a review
review.moderate	Approve/reject a review
order.create	Create a food order
order.list	List orders
order.updateStatus	Update order status
order.get	Get order details
event.create	Create an event
event.list	List events
event.update	Update an event
event.cancel	Cancel an event
loyalty.createMember	Register loyalty member
loyalty.listMembers	List members
loyalty.addPoints	Add points
loyalty.redeemPoints	Redeem points
loyalty.getMember	Get member + history
seo.getPageMeta	Get page SEO metadata
seo.updatePageMeta	Update page SEO metadata
seo.getStructuredData	Get restaurant structured data
seo.updateStructuredData	Update structured data
review.pending	List all draft items (HITL)
review.approve	Publish a draft item
review.reject	Archive a draft item
site.get_context	Get site context
site.update_context	Update site context

Cognix Skills

Generated skill packages are in skills/. Install into Cognix:

cd skills/dineagent-menu && cognix skill install .

Skills delegate to the CLI via child_process, making DineAgent components available as Cognix agent tools.

Development

Commands

pnpm build          # Build all packages
pnpm dev            # Start all dev servers
pnpm lint           # Run ESLint
pnpm format         # Format with Prettier
pnpm test           # Run tests
pnpm generate-skills # Regenerate Cognix skill packages

Adding a New Component

1. Create packages/components/<name>/ with the 6-file contract
2. Add package.json and tsconfig.json
3. Register in apps/cms/src/payload.config.ts
4. Register in apps/mcp/src/registry.ts
5. Register in apps/cli/src/index.ts
6. Add workspace deps to apps' package.json
7. Run pnpm generate-skills

Tech Stack

Layer	Technology
Frontend	Astro (static site generation)
CMS	Payload CMS 3.x (self-hosted)
Database	SQLite (dev) / PostgreSQL (prod)
AI Interface	MCP SDK (Model Context Protocol)
CLI	Commander.js
Validation	Zod
Monorepo	Turborepo + pnpm workspaces
Skills Bridge	component.yaml → skill.yaml + handler.mjs

License

MIT