360 Secure SkillHub
360 安全技能中心

mcdonalds-mcp-order-lite

该 Skill 借助远程 MCP 服务完成麦当劳中国外卖下单流程:先查询可配送地址与门店菜单,查看餐品详情、计算实付价格,经用户确认后创建订单并给出支付入口,支付后再查询订单状态;同时支持优惠券、积分商城、营养信息与营销日历查询。包内含 API 速查文档、命令行 RPC 脚本和若干 Python 辅助模块,不内置访问令牌。

danielliao
danielliao
热度 65.59
最近更新 2026-05-11 17:29:48
分类:生活依赖:MCP 服务 / 第三方 API / 代码执行 / 读写文件应用场景:购物 / 推荐 / API 集成 / 自动化脚本

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add mcdonalds-mcp-order-lite
Metadata
name
mcdonalds-mcp-order-lite
description
Place McDonald's China delivery orders through the official MCP server at https://mcp.mcd.cn using a Bearer MCP token over Streamable HTTP / JSON-RPC. Use when the user wants to browse McDonald's deliverable addresses, query store menu items, inspect meal details, calculate price, create a delivery order, query order status, or check McDonald's coupons/points. Also use when wiring this MCP into a client like Cursor, Cherry Studio, Trae, or another agent. This lite package intentionally contains no embedded token and only the core reusable files.

Use the official McDonald's China MCP toolchain, not scraped web APIs.

Core workflow

  1. Identify the delivery address.
  2. Call delivery-query-addresses first.
  3. Pick the matching address record and carry forward addressId, storeCode, and beCode from the same record.
  4. Call query-meals for that store.
  5. If needed, call query-meal-detail for a chosen meal code.
  6. Call calculate-price before ordering.
  7. Show the user the real payable amount and wait for confirmation.
  8. Call create-order only after confirmation.
  9. Return the payH5Url to the user.
  10. After the user says payment is complete, call query-order.

Important constraints

  • Use server URL https://mcp.mcd.cn.
  • Send Authorization: Bearer <TOKEN>.
  • MCP protocol version must be 2025-06-18 or earlier.
  • Prefer low request volume; the docs say 600 requests/minute/token.
  • Do not invent storeCode, beCode, or addressId.
  • For delivery ordering, always use values returned by delivery-query-addresses or delivery-create-address.
  • Always price-check with calculate-price before create-order.
  • query-meal-detail docs note that replacing default combo choices is not yet supported in current version.

Tool map

Delivery ordering

  • delivery-query-addresses: list saved deliverable addresses
  • delivery-create-address: add a new delivery address
  • query-store-coupons: coupons usable for the chosen store
  • query-meals: menu for the chosen store
  • query-meal-detail: inspect a meal / combo
  • calculate-price: final price check
  • create-order: create payable delivery order
  • query-order: query order status after creation/payment

Other useful tools

  • available-coupons: claimable coupons
  • auto-bind-coupons: one-click claim current coupons
  • query-my-coupons: coupon wallet
  • list-nutrition-foods: nutrition list
  • query-my-account: points account
  • mall-points-products, mall-product-detail, mall-create-order: points mall
  • campaign-calendar: marketing calendar
  • now-time-info: server time

Recommended user-facing flow

When the user says “帮我点麦当劳” or similar:

  • Confirm budget / preferred item if missing.
  • Resolve the delivery address from saved addresses first.
  • Find candidate items from query-meals.
  • Use calculate-price for one or a few concrete combinations.
  • Present the cheapest / best-fit options with real payable totals.
  • Only place the order after explicit confirmation.
  • After order creation, send the payment link.
  • After payment, query query-order instead of assuming success.

Files in this skill

  • Read references/api.md for the compact API cheat sheet and tested field names.
  • Use scripts/mcd_rpc.py for direct RPC calls to the remote MCP server when you need deterministic testing or CLI usage.
  • client.py contains a Python wrapper around the same remote MCP calls.
  • nlp_processor.py contains lightweight text parsing for order intent.
  • tools.py contains higher-level helper functions.

Packaging notes

Keep the skill folder lean. Do not add extra docs beyond what is needed for reuse. Remove transient files like __pycache__ before packaging.