Set Up a Restaurant with ChatGPT: Menu, QR Codes and Tables in One Conversation

Ask any restaurant software vendor where their free trials die and you will get the same answer, usually reluctantly: at the empty menu screen. The owner signs up, lands on a clean dashboard, sees Add your first category, and thinks about the eighty-item menu sitting in a laminated folder by the till. Then a table needs turning, a supplier calls, and the tab is closed. The software was never bad. It was just asking for ninety minutes of typing before it gave anything back.

That wall is the single biggest reason restaurants stay on paper menus and handwritten dockets, and it has almost nothing to do with the quality of the product behind it. It is a data-entry problem wearing a technology costume.

This is where a genuinely useful application of AI shows up — not as a chatbot bolted onto a dashboard, but as something that does the typing for you. RestaurantManage runs a public MCP server, which means the assistant you already use — ChatGPT, Claude, or any other MCP-capable client — can be given controlled permission to set your restaurant up while you describe it in plain language. You send a photo of your menu. It builds the menu, the categories, the prices, the tables and the QR codes.

This guide covers what that actually does, what it deliberately cannot do, and where it stops being useful.

Key Takeaways

• The blank menu screen is the real adoption barrier. Not price, not features. An eighty-item menu typed by hand is the reason most restaurant software trials are abandoned in week one.
• A photo is now a valid input format. Send a photograph of your printed menu in chat and the categories, dishes, sizes and prices are created — with translations into 8 languages and a picture for every dish.
• Setup and daily management, not operations. Fourteen tools cover menu, settings, tables and QR codes, plus everyday price and stock changes.
• Orders, payments, staff and customer data are not reachable. Not by policy — the tools do not exist, so no prompt can reach them.
• Standard, not proprietary. MCP with OAuth 2.1 and PKCE. Nothing to install, no API key, no vendor lock to a single assistant.
• Free. The MCP server costs nothing extra and the QR menu it creates is free forever.

Why Restaurant Software Dies at the Menu Screen

A mid-size independent restaurant carries somewhere between 60 and 140 sellable items once you count sizes and variants. Entering them properly means a name, a description, a price per size, a category, an allergen note and ideally a photograph — call it forty-five seconds per item if you are quick and know the software. That is well over an hour of concentrated desk work, in a job that does not contain hours of concentrated desk work.

Worse, it is unpaid work with delayed reward. The owner does not get a working QR menu after item one; they get it after item eighty. Every incentive points at closing the tab and trying again "next week", which is a date that famously never arrives.

Vendors have tried to solve this for years with spreadsheet importers, which trade one kind of tedium for another, and with paid onboarding, which does not scale below a certain price point. Neither addresses the actual shape of the problem: the menu already exists. It is printed, laminated and sitting on the table. It just is not in a format software could read.

What "Setting Up in Chat" Actually Means

The mechanism is worth understanding, because it is not what most people picture when they hear "AI feature". There is no chatbot inside our panel and no AI model of ours reading your data. Instead, RestaurantManage publishes a set of tools at a public address, and your assistant — your ChatGPT or Claude account — is given permission to call them on your behalf.

In practice, a first session looks like this:

1. You paste one URL into your assistant's connector settings and sign in to RestaurantManage once. If you do not have an account, you create it on the same screen — restaurant name, country, agreement. Nothing else.
2. You type: "Set up my restaurant — here's a photo of my menu" and attach a phone snapshot.
3. The assistant reads the photo itself, then confirms the important, hard-to-change choices with you: country, currency, menu language.
4. It creates the categories, dishes, sizes and prices, and reports back what it made. Translations into 8 languages and an image for each dish are generated in the background.
5. You say "add a terrace with 8 tables", and it creates the location with automatically numbered tables.
6. You say "give me the QR links" and "send me the printable QR cards", and it returns the per-table links plus a direct download for an A4 PDF, six cards to a page, ready to cut.

Elapsed time for a normal menu: minutes, most of it spent reading back what the assistant proposed rather than typing. The work that used to be an hour of forms becomes a conversation you can have standing up.

After Setup: The Boring Jobs Nobody Logs In For

The setup case is the dramatic one, but the assistant earns its place afterwards, on the small tasks that are individually trivial and collectively never done.

Prices are the obvious example. When a supplier raises costs, the correct response is to move a category of prices by a percentage. In a dashboard that is twenty edits, so it gets postponed, and the restaurant runs on stale margins for a month. In chat it is one sentence: "raise all pizza prices by 10%".

Stock is the other. A dish sells out at eight in the evening, which is precisely when nobody is going to open a laptop. Saying "the salmon is finished for tonight" from a phone hides it from every QR menu immediately, which is the difference between a guest ordering something you cannot make and a guest ordering something else.

The full day-to-day set: list the menu with current prices, update one price or many, mark items available or out of stock, add a dish, remove a dish.

The Part Most AI Integrations Get Wrong

Here is the uncomfortable question that any restaurant owner should ask before connecting an AI assistant to their business system: what happens when someone talks it into something stupid?

This is not paranoia. Language models are, by design, persuadable — that is the same property that makes them useful. Text arriving from outside can influence behaviour, and "the assistant was tricked" is a real category of incident rather than a theoretical one. Any integration that answers this question with "we told it not to" has not answered it.

The only durable answer is subtraction. A tool that does not exist cannot be called, no matter how the request is phrased. So the permission surface here was defined by what it leaves out:

• No orders. The assistant cannot see, create, change or cancel a single order. Your live service is untouchable.
• No money. No payments, no cash drawer, no subscription changes, and no access to turnover or reports.
• No people. No staff accounts, no roles, no passwords.
• No customer data. The phone book, saved addresses and call records are not exposed.
• No infrastructure. Printers and caller ID stay in the panel.
• Almost no deletion. The restaurant, its locations, its tables and its data cannot be deleted at all. The single exception is removing one menu item, and that tool repeats the dish name back and waits for confirmation.

Add daily write quotas per restaurant, so a malfunctioning loop cannot rewrite an entire menu overnight, and revocation that takes effect immediately rather than at the end of a session, and you have a surface where the worst realistic outcome is a wrong price that you fix in the next sentence.

That is a deliberately unambitious design, and it should be. The value of AI in a restaurant system is in removing data entry, not in being handed the till.

How It Works Underneath

Two standards do the work, and both are worth knowing by name because they are what stop this from being a proprietary trick.

MCP: the connection

The Model Context Protocol is an open standard for describing tools to an AI assistant. Because RestaurantManage speaks it, no assistant needs a special integration built for it, and no restaurant is locked to whichever assistant happens to be popular this year. The server also supports Dynamic Client Registration, which means a new client can register itself — there is no waiting list and no API key to request.

OAuth 2.1: the permission

Sign-in uses OAuth 2.1 with PKCE — the same mechanism that lets an app act on your behalf without ever seeing your password. Your assistant receives a scoped token, limited to setup and menu management for your restaurant alone. It cannot reach another restaurant on the platform, and another owner's assistant cannot reach yours. Before you approve, the consent screen shows two explicit lists: what will be possible, and what will never be possible.

Connecting It: About Two Minutes

Claude

1. Settings → Connectors → Add custom connector.
2. Paste https://api.restaurantmanage.com/mcp and press Connect.
3. Sign in with Google or e-mail, read the permission card, press Allow.

ChatGPT

1. Settings → Apps, enable Developer mode.
2. Create app, paste the same URL.
3. Sign in and press Allow.

Nothing is installed on your computer, and the same connection works from your phone.

Where AI Setup Does Not Help

An honest guide needs this section, because the failure cases are predictable.

Bad source material. A crisp photo of a printed menu imports well. A handwritten chalkboard, a heavily designed menu where prices float away from the dishes, or a photo taken at an angle in bad light will produce mistakes. You see the proposed result in the chat before it is saved, so the fix is a sentence — but if your menu only exists as handwriting, typing it may genuinely be faster.

Judgement calls. The assistant will not decide your pricing strategy, which dishes deserve photographs, or how your categories should be ordered for how guests actually browse. It is fast at execution and empty on taste.

Anything operational. Kitchen display setup, printer configuration, staff roles, table plans in the physical sense, payment settings — all panel work, by design. If your problem is "my kitchen tickets print in the wrong place", the assistant is not the tool.

Very large or unusual menus. Several hundred items with complex modifier logic will still want a human pass afterwards. The assistant gets you to ninety percent quickly; the last ten percent is a review, not a conversation.

What It Costs

The MCP server is included at no additional charge, and the QR menu it sets up is free forever with unlimited items and unlimited views. You need a RestaurantManage account and an AI assistant you already use — free or paid on their side, it makes no difference to us. There is no per-restaurant licence for the assistant and no setup fee.

Getting Started

If you have a printed menu and five minutes, the fastest path is not our sign-up form. Open your assistant, paste the connector URL, sign in, and send a photo of the menu. By the time you have read back what it created, you have a working QR menu you can put on a table.

For the step-by-step version, including troubleshooting and the full tool list, see the AI Assistant help guide.

Related Reading

• What is a QR Menu? How to Create One for Your Restaurant
• How to Create a QR Menu: Step-by-Step Setup Guide
• Free Restaurant Management System: QR Menu, 3 Mobile Apps & Full Operations
• Best Free QR Menu: What to Look For
• Restaurant Digital Transformation Guide

Written byUtku Berberoglu

Founder of RestaurantManage. Develops QR menu and restaurant management solutions for the food service industry.