Explainer
AI restaurant ordering starts with a menu an assistant can read
An AI assistant can build a guest's cart from your live menu, but the guest still pays on your checkout. Here is what it reads, what it can't do, and how to check your menu.

What an AI-readable menu is
An assistant can open a PDF menu. It can't tell that the brioche bun is free, that bacon costs $1.50 more, or that the fries ran out at 8 pm. An AI-readable menu is one the assistant can question and get structured answers from, store by store. Your storefront already holds that data, and what was missing was a way to ask. Analysts call software that shops for a person agentic commerce. For a restaurant the idea is narrower: one guest, one cart, one checkout.
- Structured: item names, prices, and choice groups with their rules, such as pick one bun.
- Live: a sold-out item is marked unavailable as soon as it changes.
- Per location: each store has its own hours, fulfillment modes, and menu.
- Priced like your checkout: the assistant never quotes a total your checkout won't charge.
What an MCP server is, in plain words
MCP stands for Model Context Protocol. It is a standard way for software to publish a short list of things it can do, so an assistant can call them by name. A restaurant MCP server answers two kinds of question: what is on the menu at this store, and what would this cart cost. It doesn't chat. The assistant holds the conversation and writes every reply, and the server returns facts.
- Server: one web address. Appetier gives each brand its own.
- Tool: a named question with set inputs, such as a location and a list of items.
- Assistant: the app the guest talks to. It picks which tools to call.
- Stateless: the assistant sends the whole cart on each call, and the server keeps no session.
What your server exposes: four tools
The server needs no sign-in, and the brand comes from its web address alone. A call that names a location your brand doesn't own is refused before anything is read. Four tools cover an order from first question to checkout.
- get_restaurant: your locations, their hours, and the fulfillment modes each offers.
- get_menu: one location's menu with modifiers, prices, and live availability.
- quote_cart: prices the cart with tax and fees through the same order service and checks as web checkout: store hours, prep time, and live inventory.
- create_checkout_link: stores the cart and returns a link to your own checkout.
One assistant order, four calls
Here is a pickup order at the Royal Burger demo store. The guest tells an assistant: a Royale with Cheese on a brioche bun with bacon, and a medium fries. The prices below come from the demo menu, and the assistant makes four calls.
- Call 1, get_restaurant: returns the Fairfax location, its hours, and that it takes pickup and delivery.
- Call 2, get_menu: returns Royale with Cheese at $7.09 with a required bun choice, thick-cut bacon at +$1.50, and Royal Fries (M) at $3.39 with optional seasoning.
- Call 3, quote_cart: sends both lines with the brioche bun and the bacon. The subtotal is $7.09 + $1.50 + $3.39 = $11.98, and the result adds the tax and fees the location sets.
- If a line were sold out or the store were closed, call 3 would list that as an issue, and no link would follow.
- Call 4, create_checkout_link: prices the cart once more, stores it, and returns a link to the Royal Burger checkout that lasts 30 minutes.
What the same order does when something is wrong
A good server says no clearly. The assistant relays the answer, so each refusal below is worded for a guest to read, and none of them reaches your kitchen. The pattern is the same everywhere: the server refuses before it writes anything.
- Fries sold out: the quote carries an inventory issue, and the server returns no checkout link.
- Store closed or prep time not met: the quote lists the hours issue, exactly as web checkout would.
- Another brand's location id: the call fails as not found, and no menu is read.
- A price or a card number in the arguments: the call is refused as invalid, nothing is stored, and the value isn't logged.
- Too many checkout links from one assistant user: the call is refused, and no link record is written.
- Assistant ordering switched off: every tool answers that the restaurant doesn't take orders through AI assistants, and names your storefront.
How the order ends on your own checkout
The assistant never places the order. The guest opens the link, and your storefront loads the cart and prices it again before payment. Contact details and card are typed there, on your page. Card handling stays out of the assistant, so Appetier stays at the lightest card-security level, SAQ A.
- An item that sold out since the link was made shows the normal sold-out message before payment.
- An expired or used link opens your menu with a note that the order link expired.
- A link serves one order. A second order from the same link is a normal web order.
- No tool accepts a name, an email, a phone number, an address, or a card field.
What you see in the console, and the off switch
An order from a link arrives on its own channel, labelled assistant, with the host that made the link. Console orders and analytics show it apart from web, mobile, and marketplace orders, so you can see what the channel brings. It counts toward the Growth fee like any direct order, with no commission.
- It is on for every brand. One switch on the console's Integrations page turns it off, and then every tool refuses and points the guest to your storefront.
- A link made before you switch it off still works, because the guest is already on your page.
- Assistant data is kept 30 days at most, and a checkout link record lasts 30 minutes. The order itself is kept like any web order.
- Appetier runs no AI model in this path. The assistant's own service does the talking, and we answer menu and price questions.
Where AI restaurant ordering stops
Appetier isn't listed in the assistant directories yet, so guests won't find your restaurant by searching inside an assistant app. You share your server link instead. A guest adds it to an assistant that takes custom connectors. Copy the link from the Integrations page of the console, then put it on your site, your receipts, and your social profiles.
- Payment: the guest pays on your checkout page. That is one extra tap, and the price of keeping cards out of the assistant.
- Sign-in: none. There is no loyalty, saved preference, or order history inside the assistant.
- Codes: a coupon, reward, or gift card code can't go in through a tool. Offers your store already runs still apply.
- Prices: a caller can't set a price. A call that sends one is refused.
- Wording: the assistant writes its own replies, so its phrasing is its own. The numbers come from your menu.
AI restaurant ordering FAQs
- What is an AI-readable menu?: A menu an assistant can query for items, choices, prices, and what is sold out now. A PDF or a photo isn't one.
- What is an MCP server for a restaurant?: A web address that answers an assistant's questions about your menu and carts. Appetier gives each brand one, with four tools.
- Does the assistant take payment?: No. The guest pays on your own checkout page, and no tool accepts a card.
- Can I turn AI restaurant ordering off?: Yes. One switch on the Integrations page makes every tool refuse and point guests to your storefront.
- Does it cost extra?: No separate fee. An assistant order counts toward the Growth fee like any direct order.
- Do I need a POS?: No. Connecting a POS adds automatic price and sold-out updates.
Next steps
Work through the checklist on your own menu. Start with required choices and sold-out items. See how POS sync keeps both matched to the register, and read how guests can also speak an order in voice ordering for restaurants. To score your site and listing first, run the free audit.
See your own storefront today
Upload your logo and a photo of your menu, and your storefront goes live in minutes. Nothing is charged until direct orders come in.
