Grubhub外卖配送MCP服务

一个MCP服务器,提供Grubhub外卖平台的编程访问接口,可搜索餐厅、浏览菜单、管理购物车、下单并追踪配送,适用于将外卖点餐能力集成到AI助手或自动化工作流的场景。

  • by aserper
  • Python
  • 本地部署
  • 更新于
在 GitHub 查看 aserper/grubhub-mcp

使用说明

Grubhub MCP Server

An MCP (Model Context Protocol) server that provides programmatic access to Grubhub's food delivery platform. Search restaurants, browse menus, manage your cart, place orders, and track deliveries — all through MCP tools.

Features

  • Search & Discovery — Find restaurants by location, cuisine, or keyword with autocomplete
  • Restaurant & Menus — View restaurant details, full menus, item options and pricing
  • Cart Management — Create carts, add/remove items, apply promo codes, set tips
  • Order Flow — Place orders, track deliveries in real-time, reorder past meals
  • Account Management — Manage profile, saved addresses, favorites, and passwords
  • Payments — View saved payment methods, check gift card balances

All 36 Tools

Category Tools
Auth login, logout, get_session_info, send_login_otp, verify_login_otp, create_account, send_password_reset
Search search_restaurants, autocomplete_search
Restaurant get_restaurant, get_menu, get_menu_item
Cart create_cart, get_cart, add_to_cart, update_cart_item, remove_from_cart, apply_promo_code, set_tip
Order place_order, get_order, get_order_history, track_order, reorder, post_delivery_tip
Account get_profile, update_profile, get_addresses, add_address, get_favorites, add_favorite, remove_favorite, change_password
Payments get_payment_methods, get_gift_card_balance, apply_gift_card

Installation

Claude Code Plugin (Recommended)

claude plugin marketplace add aserper/grubhub-mcp
claude plugin install grubhub-mcp@grubhub-marketplace

uvx (No Installation Required)

Add to your Claude Code MCP settings (~/.claude/settings.json):

{
  "mcpServers": {
    "grubhub": {
      "command": "uvx",
      "args": ["--from", "grubhub-mcp", "grubhub"]
    }
  }
}

Manual / From Source

Requires Python 3.11+.

git clone https://github.com/aserper/grubhub-mcp.git
cd grubhub-mcp
uv venv && source .venv/bin/activate
uv pip install -e .

Then add to ~/.claude/settings.json:

{
  "mcpServers": {
    "grubhub": {
      "command": "/path/to/grubhub-mcp/.venv/bin/grubhub"
    }
  }
}

Usage with Other MCP Clients

Run the server over stdio:

# With uvx (no install needed)
uvx --from grubhub-mcp grubhub

# Or from a local install
grubhub

# Or as a Python module
python -m grubhub_mcp

Authentication

No login required for browsing — search restaurants, view menus, and check prices without an account. The server automatically creates an anonymous session.

Login required for ordering, account management, and order history:

Use the login tool with my email and password

Supports email/password login and OTP (one-time passcode) authentication.

Sessions are stored in ~/.grubhub-mcp/session.json (override the directory with GRUBHUB_SESSION_DIR). Access tokens are refreshed on use, including before expiry when Grubhub supplies expiry metadata and once after an authenticated request returns 401. Rotated tokens are saved for the next server invocation. Refresh responses that omit account metadata do not erase your login identity. Concurrent refreshes are serialized within the client, and on POSIX systems across server processes sharing the session directory. Session writes are atomic and use private file permissions. Grubhub's lifetime fields are in minutes; the observed access/refresh lifetimes are one hour / 30 days, but server policy may change. There is no background keep-alive: a revoked or expired refresh token still requires a new login. Do not share your session file; it contains credentials.

Order History Pagination

get_order_history uses server-side pagination so older orders are reachable. page_num is 1-based (default 1), and page_size defaults to 20. Iterate from 1 through pagination.total_pages to retrieve the full history. Both values must be positive. Order-detail and reorder fallbacks search across all available history pages.

Migration from v1.1.6: callers using the old 0-based page_num must add one to their page index. The response now provides total_pages and current_page instead of the legacy total_orders value, which only counted a capped subset.

How It Works

The API endpoints were reverse-engineered from the Grubhub Android app (v2026.11.1) by decompiling the APK with jadx and analyzing the Retrofit service interfaces, OkHttp interceptors, and data models.

Key technical details:

  • Base API: https://api-gtm.grubhub.com
  • Auth: Bearer token via email/password or anonymous sessions
  • Transport: REST/JSON for most endpoints, Protobuf for some BFF services
  • Headers: Mimics the Android app's request headers including x-gh-browser-id

Project Structure

src/grubhub_mcp/
├── __init__.py
├── __main__.py        # python -m entry point
├── server.py          # MCP server setup and tool registration
├── client.py          # HTTP client with auth and header management
├── auth.py            # Authentication flows (login, anonymous, OTP, refresh)
└── tools/
    ├── auth.py        # Auth tools (login, logout, OTP, account creation)
    ├── search.py      # Restaurant search and autocomplete
    ├── restaurant.py  # Restaurant details, menus, menu items
    ├── cart.py        # Cart CRUD, promo codes, tips
    ├── order.py       # Place orders, track, reorder, order history
    ├── account.py     # Profile, addresses, favorites, password
    └── payments.py    # Payment methods, gift cards

Development

Run the regression suite with:

uv run python -m unittest discover -s tests -v

CI runs tests on pushes and pull requests, and the release job runs them before publishing. The package currently requires MCP 1.x (mcp<2).

Disclaimer

This project is for educational and personal use. It is not affiliated with or endorsed by Grubhub. Use responsibly and in accordance with Grubhub's terms of service.

License

MIT