# 🔌 Restaurant MCP Server

Standalone MCP (Model Context Protocol) server exposing restaurant tools.
Any MCP-compatible AI agent can connect via SSE URL.

## Quick Start

```bash
npm install
cp .env.example .env
# Fill in DATABASE_URL (same Neon project as Project 3)

npm run dev   # → http://localhost:3010
```

## Tools Exposed

| Tool | Description |
|---|---|
| search_restaurants | Search by cuisine, price, rating, open now |
| get_restaurant_detail | Full details for one restaurant |
| get_user_preferences | Load saved user preferences from Neon DB |
| save_user_preferences | Save user likes, dietary needs, budget |
| search_knowledge | RAG search on cuisine and area knowledge base |

## Connect to This Server

Any MCP-compatible AI:
```js
mcp_servers: [{
  type: "url",
  url:  "http://localhost:3010/sse",
  name: "restaurant-tools"
}]
```

## Data

50 mock restaurants covering Scottsdale and Phoenix:
  - South Indian / Tamil
  - Mexican, Italian, Japanese
  - American, Thai, Mediterranean
  - Middle Eastern, Chinese
  - Breakfast, BBQ, Vegan, Seafood

User preferences stored in Neon Postgres.
Knowledge base: 10 docs covering cuisine info, dietary guidance, area tips.

## Zod Validation

All tool inputs validated with Zod schemas before execution.
Invalid inputs return structured error messages.