棒棒玛特 MCP Server

English | 中文

---

棒棒玛特 MCP

棒棒玛特（BangBangMart）的 MCP（模型上下文协议）服务器 —— 专为 AI 智能体交互而设计的在线零食商店。

MCP 服务器已上线：https://www.bangbangmart.com/mcp（Streamable HTTP，无状态，无需安装）

简介

棒棒玛特是一个 AI 优先设计的电商站点：所有接口均返回纯文本并包裹在极简 HTML 中，便于大模型直接消费。MCP 服务器以结构化工具的形式暴露了相同的商品目录、下单和订单查询功能。

• 站点主页：https://www.bangbangmart.com
• MCP 接口：https://www.bangbangmart.com/mcp
• 协议：MCP Streamable HTTP（无状态，每次请求独立 transport）
• 服务器名称：bangbangmart

工具列表

工具	说明
search_products	按关键词、价格、是否促销搜索零食目录，返回商品 ID、价格和购买链接
get_product	通过 product_id（如 p_snacks_0001）获取单个商品的完整详情
checkout	为一个或多个商品创建订单并获取支付链接，需要 viewer token
get_order_status	通过 order_id（格式：ord_XXXXXXX）查询订单状态
get_orders	列出当前用户的所有订单，需要 viewer token
pay_order	确认模拟支付

工具参数详情

checkout 支持两种下单方式：

• 单商品：传 product_id（如 p_snacks_0001）和 qty
• 多商品：传 items 字符串，格式为逗号分隔的 product_id:qty 对，如 p_snacks_0001:2,p_snacks_0014:1

Viewer Token

checkout 和 get_orders 需要 viewer token。Token 是由棒棒玛特后端签发的短 base64url 字符串（如 AABSu3k4oEw）。请让用户提供其 token —— 不要猜测或自行生成。

MCP 客户端配置

服务器已托管，无需本地安装，直接将 MCP 客户端指向远程接口即可。

Cursor

{
  "mcpServers": {
    "bangbangmart": {
      "type": "http",
      "url": "https://www.bangbangmart.com/mcp"
    }
  }
}

Claude Code

claude mcp add --transport http bangbangmart https://www.bangbangmart.com/mcp

Visual Studio Code

{
  "servers": {
    "bangbangmart": {
      "type": "http",
      "url": "https://www.bangbangmart.com/mcp"
    }
  }
}

Kimi Code

kimi mcp add --transport http bangbangmart https://www.bangbangmart.com/mcp

智能体工作流示例

1. 调用 search_products（如 q: "薯片"）查找商品及其 ID
2. 调用 get_product 传入 product_id 查看完整详情
3. 请用户提供其 viewer token
4. 调用 checkout 传入 token、product_id 和 qty（或多商品 items）下单
5. 将响应中的支付链接展示给用户
6. 调用 pay_order 传入 order_id 确认模拟支付
7. 调用 get_order_status 或 get_orders 验证订单状态

许可证

MIT — 详见 LICENSE

---

BangBangMart MCP

MCP (Model Context Protocol) server for 棒棒玛特 (BangBangMart) — an online snack shop designed for AI agent interaction.

The MCP server is live at https://www.bangbangmart.com/mcp (Streamable HTTP, stateless, no installation required).

Overview

BangBangMart is an e-commerce site built AI-first: all routes return plain text wrapped in minimal HTML, making them easy for LLMs to consume directly. The MCP server exposes the same catalog, checkout, and order functionality as structured tools.

• Live site: https://www.bangbangmart.com
• MCP endpoint: https://www.bangbangmart.com/mcp
• Protocol: MCP Streamable HTTP (stateless, one transport per request)
• Server name: bangbangmart

Tools

Tool	Description
search_products	Search and filter the snack catalog by keyword, price, sale status. Returns product IDs, prices, and buy links.
get_product	Get full details for a single product by product_id (e.g. p_snacks_0001).
checkout	Create an order for one or more products and get a payment URL. Requires a viewer token.
get_order_status	Look up the status of an order by order_id (format: ord_XXXXXXX).
get_orders	List all orders for the authenticated viewer. Requires a viewer token.
pay_order	Confirm mock payment for an order.

Tool parameter details

checkout supports two ordering modes:

• Single product: pass product_id (e.g. p_snacks_0001) and qty
• Multiple products: pass an items string of comma-separated product_id:qty pairs, e.g. p_snacks_0001:2,p_snacks_0014:1

Viewer tokens

checkout and get_orders require a viewer token. Tokens are short base64url strings (e.g. AABSu3k4oEw) issued by the BangBangMart backend. Ask the user to provide their token — do not generate or guess one.

MCP client configuration

Because the server is hosted, no local installation is needed. Configure your MCP client to point at the remote endpoint.

Cursor

{
  "mcpServers": {
    "bangbangmart": {
      "type": "http",
      "url": "https://www.bangbangmart.com/mcp"
    }
  }
}

Claude Code

claude mcp add --transport http bangbangmart https://www.bangbangmart.com/mcp

Visual Studio Code

{
  "servers": {
    "bangbangmart": {
      "type": "http",
      "url": "https://www.bangbangmart.com/mcp"
    }
  }
}

Kimi Code

kimi mcp add --transport http bangbangmart https://www.bangbangmart.com/mcp

Example agent workflow

1. Call search_products with a keyword (e.g. q: "薯片") to find products and their IDs.
2. Call get_product with a product_id to see full details.
3. Ask the user for their viewer token.
4. Call checkout with the token, product_id, and qty (or multi-product items) to place an order.
5. Present the payment URL from the response to the user.
6. Call pay_order with the order_id to confirm mock payment.
7. Call get_order_status or get_orders to verify.

License

MIT — see LICENSE.