FoodDash
An MCP server and app that enables users to browse restaurants, view menus, manage shopping carts, and place food orders with live status tracking. It serves as a reference implementation for MCP Apps SDK patterns like tool visibility, lifecycle hooks, and structured content.
README
FoodDash
A food ordering MCP App that demonstrates key MCP Apps SDK patterns. Browse restaurants, view menus, add items to cart, and place orders — all within an MCP host.
Getting Started
bun install
bun run dev
The server starts at http://localhost:3001/mcp.
Scripts
| Command | Description |
|---|---|
bun run dev |
Start dev server with hot reload (web + server watch) |
bun run build |
Production build (typecheck → web → server bundle) |
bun run start |
Run the built server in production mode |
bun run typecheck |
Type-check only |
bun run build:web |
Build web assets only |
bun run build:server |
Bundle server entry only |
bun run dev:server |
Watch and restart server |
MCP Apps Concepts Demonstrated
This app is designed as a learning reference. Each concept is documented with inline comments in the source code.
Tool Visibility
Two kinds of tools share one UI:
| Tool | Visibility | Purpose |
|---|---|---|
order-food |
Model + App | Model calls this when user says "order food". Opens the UI. |
get-order-summary |
Model + App | Model calls this when user asks "where's my food?". Same UI. |
get-menu |
App only | UI calls this when user taps a restaurant. Model never sees it. |
place-order |
App only | UI calls this when user clicks "Place Order". |
get-order-status |
App only | UI polls this every 10s for live order tracking. |
apply-promo |
App only | UI calls this when user applies a promo code. |
Why it matters: Model-visible tools give the LLM a clean, high-level interface ("order food", "check order status"). App-only tools handle granular CRUD that shouldn't clutter the model's tool list.
Lifecycle Hooks (mcp-app.tsx)
| Hook | When it fires | What FoodDash does |
|---|---|---|
ontoolinput |
Tool invoked, BEFORE server responds | Pre-sets cuisine filter from model's cuisine argument |
ontoolresult |
AFTER server returns | Loads restaurant list or order data into state |
ontoolcancelled |
Host cancels a tool call | Resets loading spinners |
onteardown |
Host is closing the app | Returns order summary or abandoned cart info to the model |
onhostcontextchanged |
Host theme/viewport changes | Updates safe area insets |
Structured Content
Every tool result carries two payloads:
content[]— text the model reads in its conversationstructuredContent— JSON the app UI parses to render
They can carry different levels of detail. See server.ts for examples.
Shared Resource URI
Both order-food and get-order-summary point to the same resourceUri. The host renders one UI instance and routes different tool results to it. The app distinguishes them by the shape of structuredContent.
App → Server Communication
The app calls server tools via app.callServerTool(). This is the reverse of the model calling tools — the app initiates calls for app-only operations like fetching menus and placing orders.
Features
- Browse 5 restaurants with ratings, hours, delivery info, and feature tags
- Search, filter by cuisine, and sort by rating/delivery time/fee
- Menu items with dietary labels, calorie counts, and popularity badges
- Dietary filter pills on the menu view
- Cart with quantity controls, special instructions, and promo codes
- Promo codes:
WELCOME10(10% off),FREEDEL(free delivery),SAVE5($5 off) - Live order tracking with auto-advancing status (confirmed → preparing → on the way → delivered)
- Light and dark mode support
Project Structure
server.ts MCP server, tools, and mock data
main.ts Express entry point (HTTP + stdio)
src/
mcp-app.tsx React entry point + lifecycle hooks
types.ts Shared TypeScript interfaces
constants.ts Dietary labels, order steps, sort options
global.css Design tokens and animations
hooks/
useAppState.ts Central state management + server tool calls
components/
Header.tsx Sticky header with back button and cart badge
DeliveryAddressBar.tsx Address input
RestaurantsView.tsx Restaurant list with search/filter/sort
RestaurantCard.tsx Restaurant card
SortDropdown.tsx Sort selector
RestaurantBanner.tsx Restaurant detail header in menu view
MenuView.tsx Menu with dietary filters and categorized items
MenuItemCard.tsx Menu item with dietary badges and calories
DietaryBadge.tsx Dietary label pill
CartView.tsx Cart, promo code, special instructions, summary
QuantityControl.tsx +/- quantity buttons
OrderTrackingView.tsx Live order tracking with progress stepper
ProgressStepper.tsx 4-step horizontal progress indicator
EmptyState.tsx Empty state placeholder
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。