MCP A2A AP2 Food Delivery & Payments
Enables AI agents to discover and order food from multiple delivery services (DoorDash, UberEats, Grubhub) using A2A protocol and process payments via Stripe with AP2 protocol mandates for cryptographically signed user authorization.
README
mcp-a2a-ap2-im-hungry 🍔💳
MCP server with A2A protocol (food delivery) + AP2 protocol (payments with Mandates).
Demo

Flow

Quick Start
git clone https://github.com/tas1337/mcp-a2a-ap2-im-hungry.git
cd mcp-a2a-ap2-im-hungry
npm start
Services:
- MCP Server:
http://localhost:8000/mcp - Registry:
http://localhost:8004 - DoorDash/UberEats/Grubhub:
8001-8003(A2A) - Stripe:
http://localhost:8005(AP2)
What's Different from mcp-a2a-im-hungry?
| mcp-a2a-im-hungry | This project |
|---|---|
| A2A only | A2A + AP2 |
| No payments | Stripe payments |
| — | Mandates (signed authorization) |
What is AP2?
Problem: AI agents spending money. Who approved it?
AP2 = Payments with Mandates — cryptographically signed proof of user intent.
MANDATE = { userId, maxAmount, expires, signature }
Flow (matches real-world AP2):
- User asks for cheeseburger 🍔
- Your agent searches restaurants via A2A (discovers DoorDash/UberEats/Grubhub agents)
- Your agent gets menus and delivery estimates
- Your agent decides which restaurant (cheapest/fastest)
- User authorizes payment → User's authorization service creates mandate (signed with user's private key)
- Your agent requests authorization → Receives mandate from user
- Your agent sends payment request WITH mandate to Stripe agent (via A2A)
- Stripe agent VERIFIES mandate (signature valid? not expired? amount ok?)
- If valid → Stripe processes payment ✅
- Your agent places order with food delivery agent (DoorDash/UberEats/Grubhub via A2A)
- 🍔 Burger incoming!
Key Point: The agent NEVER creates mandates. User authorization creates them. Agent only uses them.
Project Structure
src/
├── a2a-client.ts # Calls ANY agent (food or Stripe)
├── tools.ts # Food tools
├── ap2-tools.ts # Payment tools (requests user authorization)
├── user-authorization.ts # User creates mandates (real-world: wallet/auth service)
mock-agents/
├── food-delivery/ # DoorDash, UberEats, Grubhub
├── payments/ # Stripe (verifies mandates!)
└── registry/ # Agent discovery
Tools
| Tool | Description |
|---|---|
search_restaurants |
Search food (A2A) |
get_menu |
Get menu (A2A) |
place_order |
Order food (A2A) |
process_payment |
Pay via Stripe (AP2 + Mandate) |
process_refund |
Refund via Stripe (AP2 + Mandate) |
Connect to Cursor
{
"mcpServers": {
"mcp-a2a-ap2-im-hungry": {
"url": "http://localhost:8000/mcp"
}
}
}
Resources
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。