myhotlunchbox-mcp
Enables reading the school lunch calendar, managing students, placing and changing orders, and tracking deliveries and payments on a My Hot Lunchbox parent account. It signs in server-side with account credentials and exposes 38 MCP tools with confirm-gated write operations.
README
myhotlunchbox-mcp
MCP server for My Hot Lunchbox — read the school lunch calendar, manage students, place and change orders, and track deliveries and payments on a parent account.
Developed and maintained by AI (Claude Code). Use at your own discretion.
Install
npx myhotlunchbox-mcp
Or as a Claude Code plugin:
/plugin marketplace add chrischall/myhotlunchbox-mcp
/plugin install myhotlunchbox-mcp
Configure
MYHOTLUNCHBOX_USERNAME=you@example.com
MYHOTLUNCHBOX_PASSWORD=…
That is the whole setup. The server performs a real server-side sign-in against
ordernow.myhotlunchbox.com (OAuth2 password grant) and renews the session with
the refresh token it receives — no browser extension, no signed-in tab, no
captured cookie. Nothing is written to disk.
MYHOTLUNCHBOX_BASE_URL overrides the app origin if it ever moves.
The server boots without credentials so a host's install-time tools/list probe
still works; the configuration error surfaces on the first tool call.
Tools
34 tools, all prefixed mhlb_. All 20 read tools are verified live against a real parent account (node scripts/verify-reads.mjs); the 14 write tools are not — see below.
Account — mhlb_whoami, mhlb_session_reset
Students — mhlb_list_students, mhlb_get_student_form,
mhlb_new_student_form, mhlb_create_student, mhlb_update_student,
mhlb_delete_student
Calendar — mhlb_get_calendar, mhlb_get_day
Ordering — mhlb_get_cart, mhlb_get_cart_tabs, mhlb_get_menu,
mhlb_get_order_form, mhlb_get_order, mhlb_create_order,
mhlb_update_order, mhlb_delete_order
Billing — mhlb_list_transactions, mhlb_get_transaction,
mhlb_list_subscriptions, mhlb_get_subscription_settings,
mhlb_set_subscription_enabled, mhlb_unsubscribe_order,
mhlb_list_gift_cards, mhlb_apply_gift_card, mhlb_get_coupon,
mhlb_apply_coupon, mhlb_remove_coupon
Checkout — mhlb_init_checkout, mhlb_checkout
Reports — mhlb_print_calendar, mhlb_print_orders,
mhlb_print_transaction. These return real PDFs; each writes the file and
returns its path, or the bytes inline with inline: true. Set
MYHOTLUNCHBOX_OUTPUT_DIR to choose where they land (defaults to the working
directory); existing files are never overwritten.
Writes are confirm-gated
Every mutating tool takes confirm. Without confirm: true it makes no
network call and returns a dry-run preview of exactly what it would send.
mhlb_checkout charges a real payment method. The server prices the charge from
orderIds, so nothing client-side can bind the amount — there is no total in the
request to check against. expectedTotal is therefore attribution, not a
guard: you state what you expected, and it is recorded in the dry run and in
the result so an unexpected charge is traceable to the call that made it. What
the tool does refuse outright is paying a non-zero total with no orderIds.
Writes: shapes captured, acceptance unverified
npm run capture:writes runs every mutating tool against a local proxy that
forwards reads to the real service but answers writes itself, so the payloads
are built from genuine server models and nothing happens upstream. It also
proves all 13 refuse to send anything without confirm: true.
What that established, and corrected: mhlb_delete_order and
mhlb_unsubscribe_order take {orderId, eventDate, studentId, isRepeated, isSubscribed} — not the order model — and checkout takes
{orderIds, checkoutType, couponCode, giftCardCode, schoolDonations}.
What is still unverified is whether the server accepts these bodies. Shape
is not acceptance; only a real write shows that, and none has been made. Inspect
the dry-run preview before confirming, and re-read afterwards — a 200 is not
proof a write persisted.
Two limits on mhlb_checkout specifically:
- It can only pay with a card already saved on the account. Paying with a new card needs a Stripe token minted by Stripe.js in a browser, which no server-side client can produce.
- It generates an idempotency key and returns it. If a checkout fails
ambiguously, retry with that same
idempotencyKeyrather than a fresh call — that is what stops a retry becoming a second charge.
Ordering is read-modify-write
There is no "add item X" call. Fetch the model, edit it, send it back whole:
mhlb_get_menu— what is orderable for a student on a datemhlb_get_order_form— the order model to fill inmhlb_create_order— send it back (withconfirm: true)mhlb_init_checkout→mhlb_checkout— price, then pay
Fields omitted from the payload are cleared, not preserved.
Shell skill
skills/myhotlunchbox covers the same account from a shell with curl — no MCP
process needed. Useful in scripts, or on a machine where this server is not
installed.
Notes
/deliveryInfo/*and/calendar/viewMatchedVendorslook parent-facing in the compiled client but return403for a parent account — they belong to the school/vendor dashboards. No tool wraps them.- Only the parent role is wired. The same API also serves school-admin and
vendor roles; those endpoints return
403, which the client reports as a role mismatch rather than a broken session. docs/MYHOTLUNCHBOX-API.mdrecords how the API was mapped and exactly what is verified.docs/api-surface.txtis the full 359-endpoint extraction.
Licence
MIT
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。