Can ChatGPT Create a Restaurant Menu? Yes, and Now It Can Publish One Straight to Your Tables
The Duckhub team builds QR menu and online ordering software, run through your own AI assistant (ChatGPT, Claude), used by cafes, bars, and restaurants. We write practical guides based on what we see working across thousands of published menus.
Yes, ChatGPT can create a restaurant menu. Out of the box it drafts the content: categories, dish descriptions, prices, translations. Connected to a menu platform, it goes further and builds the real thing: a published, scannable QR menu with photos and working links. The difference is one connection, and it takes about two minutes to set up.
TL;DR
- Free ChatGPT handles menu content well: structure, descriptions, translations, pricing suggestions.
- Alone, it cannot host a menu: no permanent URL, no QR code that stays current, no way to mark items sold out.
- Connected to a menu platform through the Duckhub app in ChatGPT’s Plugins directory, ChatGPT creates and publishes the live menu itself.
- Claude does the same via a custom connector; Claude Code, Cursor, and Codex connect with an API key.
- Always review AI output for prices and allergens before publishing.
What can ChatGPT do for a menu without any setup?
ChatGPT with no configuration is a strong menu copywriter and organizer. Paste your dish list and it groups items into sensible categories, writes short appetizing descriptions, suggests price formatting, flags gaps (“you have no vegetarian mains”), and translates the whole thing into another language. For a brand-new venue it will also brainstorm the menu itself from your cuisine and concept.
With roughly 800 million people using ChatGPT weekly (a figure OpenAI shared in October 2025), this is already how many owners draft their first menu. The output is text, though, and text is only half the job. Somebody still has to put it somewhere guests can scan.
What can’t ChatGPT do for a menu on its own?
ChatGPT alone cannot give your menu a home. It has no way to host a web page, so there is no URL for a QR code to point at, no dashboard to change a price on Friday night, and no sold-out toggle. Everything it produces must be manually copied into whatever actually serves the menu to guests.
That manual transfer is where the friction lives: every update means another conversation, another copy-paste, another chance for the printed QR code and the real menu to drift apart. The fix is not a better prompt. It is giving the assistant hands.
How does ChatGPT build a real menu through MCP?
MCP (Model Context Protocol) is the open standard, released by Anthropic in November 2024 and since adopted across the industry, that lets AI assistants operate external services through typed tools. Duck Hub MCP allows ChatGPT, Claude and other AI assistants to create a complete restaurant menu directly inside Duck Hub: real products and categories in a real account, not text in a chat window.
The setup, in guest-facing terms:
- Create the account and the venue first. Sign up at Duckhub on the free Egg plan and add your restaurant. This step comes before ChatGPT for a concrete reason: when you connect the app, you choose which venue the assistant may manage, and an account with no venue has nothing to choose. It takes under a minute and needs no card.
- Add the Duckhub app in ChatGPT. Open the Plugins directory, search for “Duckhub QR Menu”, open it and select Connect, then sign in with your Duckhub account (Google or email) and approve the consent screen. There is no API key and no URL to paste: the OAuth flow handles it. OpenAI replaced its App Directory with the Plugin Directory on 9 July 2026, so plugins are now where connected apps live across ChatGPT web, desktop, Work and Codex, and the directory is available across ChatGPT plans. We added and ran the app on a standard free account; if you are inside a company Business or Enterprise workspace, an administrator decides which plugins staff can install.
- Ask for the menu. Paste a dish list, or upload a photo or PDF of an existing menu: “Build my full menu in Duckhub from this file: categories, dishes, prices, and a one-line description each.”
- Watch it work. The assistant calls the platform’s tools (39 of them, covering products, images, translations, promotions, publishing) and the platform validates every call, so malformed data is rejected rather than saved.
- Review, then say “publish.” Nothing is guest-visible until the menu is published, and publishing keeps a rollback snapshot. The result is a live QR menu at your venue’s own URL.
One detail worth knowing before you connect a second assistant. The connection is bound to one venue at a time, and every connected AI app shares that same choice. Connect ChatGPT to venue A and later connect Claude to venue B, and both assistants now point at venue B — the consent screen warns you when a connection is about to switch. Operators running two venues from AI should use a dk_live_ API key per venue instead, which keeps them separate.
The full step-by-step, including the built-in-AI and free-chatbot alternatives, is in our guide to creating a restaurant menu with AI.
Can Claude create a restaurant menu too?
Yes. Claude connects to the same menu platform the same way: in claude.ai, add a custom connector with the URL https://mcp.duck-hub.com/mcp, sign in, and Claude gets the identical toolset ChatGPT does. Custom connectors are available on every Claude plan, including the free tier (which allows one custom connector).
The developer-flavored variants work as well: Claude Code, Cursor, and Codex connect to the same MCP endpoint with a dk_live_ API key instead of OAuth. In practice owners use whichever assistant they already pay for; the menu tools and the results are the same. Restaurants that want an agent to analyze the menu but never touch it can issue a read-only key, which the platform enforces server-side.
ChatGPT alone vs ChatGPT connected: what actually changes?
| ChatGPT alone | ChatGPT + menu platform (MCP) | |
|---|---|---|
| Menu text (descriptions, structure) | Yes | Yes |
| Reads your PDF or menu photo | Yes | Yes, and turns it into real menu items |
| Creates a live, scannable QR menu | No | Yes |
| Updates prices later | New chat + manual copy | “Raise all coffee prices 5%” — done |
| Marks items sold out | No | Yes |
| Translations published as menu languages | Copy-paste per language | Directly on the menu |
| Cost | Free | Free (Duckhub Egg plan + free ChatGPT both work) |
Bottom line: use ChatGPT alone to think; connect it to a platform to ship. The connected mode is what turns “AI wrote my menu” into “AI runs my menu.”
What can you actually ask it once connected?
Once the app is connected, the useful prompts stop looking like copywriting requests and start looking like instructions to a staff member. The assistant is operating your real account, so each of these maps to actual dashboard actions.
Building from what you already have:
- “Here’s a photo of our current menu. Recreate it in Duckhub with the same categories and prices, and write a one-line description for each dish.”
- “Import this PDF, but skip the drinks section — we’re rebuilding that separately.”
- “Set up three categories: Breakfast, All Day, and Sunday Roast. Move every egg dish into Breakfast.”
Running it day to day:
- “Mark the sea bass sold out.” (The change reaches every table before the next guest scans.)
- “Raise all coffee prices by 40 cents, but leave the filter coffee alone.”
- “Add today’s special: pumpkin risotto, £16, vegetarian, with a description.”
- “Translate the whole menu into German and Polish, and add them as menu languages.”
- “Hide the summer cocktails category until May.”
Asking questions rather than giving orders:
- “Which of my starters have no photo?”
- “What are my ten most expensive items, and how many of them are mains?”
- “Do any dish descriptions claim something is gluten-free?” — a genuinely useful audit, since that is the claim most worth verifying by hand.
The last group matters more than it looks. A restaurant that wants an assistant to analyse the menu but never change it can issue a read-only API key instead of the full connection, and the platform enforces that server-side rather than trusting the assistant to behave.
Two limits worth setting expectations on. The assistant cannot invent photographs of your actual food: it can generate images, but a generated dish is not the dish that arrives at the table, and guests notice the difference faster than they mention it. And it will not publish anything until you say so, which is the point rather than a limitation — the draft state is exactly where the review below happens, and skipping it is how a transposed price reaches a guest.
What should you double-check in an AI-built menu?
Check prices, allergens, and anything the AI extracted from a photo or PDF. Language models are excellent at formatting and phrasing and merely good at reading numbers out of images; a transposed digit in a price costs real money, and an invented “gluten-free” label is a safety problem, not a style problem.
A sensible review pass takes five minutes: read every price against your source, strip any dietary claim you did not explicitly provide, and taste-check a few descriptions for invented ingredients. The draft-then-publish flow exists precisely so this pass happens before guests see anything.
Try it now: create a free Duckhub account, then connect ChatGPT or Claude and ask it to build your menu. Free plan, 0% commission, rollback included.