A QR menu app is now in ChatGPT's Plugins directory, so restaurants can edit menus by chatting
The Duckhub team builds AI-powered QR menu and online ordering software used by cafes, bars, and restaurants. We write practical guides based on what we see working across thousands of published menus.
The Duckhub QR Menu app was listed in ChatGPT’s Plugins directory in August 2026, giving restaurant operators a way to build, translate, reprice and publish a live QR menu without leaving a chat window. Once the app is connected to a Duckhub account, instructions like “mark the sea bass sold out” or “translate the menu into German” become real changes on the menu guests scan, rather than text the operator has to copy somewhere else.
Disclosure: Duckhub is our own product, so treat the description of what the app does as vendor information. The part of this story worth attention is not the listing itself. It is what the listing is an instance of: restaurant software crossing the line from something an AI assistant can describe to something an AI assistant can operate.
What the app actually does
The connection exposes 39 typed tools covering products, categories, images, translations, promotions, publishing and orders. In practice that means an operator can hand ChatGPT a photo or PDF of an existing menu and ask for the whole thing to be rebuilt as a structured digital menu, then keep running it conversationally afterwards.
The ongoing half matters more than the setup. Building a menu happens once; changing it happens weekly. A connected assistant handles the changes that normally mean opening a dashboard on a Friday night: hiding a category, adding a special, raising a subset of prices, publishing a new language.
Nothing reaches guests automatically. Changes stay in a draft state until the operator asks for them to be published, and publishing keeps a rollback snapshot. An operator who wants analysis without write access can issue a read-only key instead, enforced server-side rather than by asking the assistant to behave.
How an operator connects it
- Create the Duckhub account and add the restaurant. This comes first because the connection asks which venue the assistant may manage, and an empty account has nothing to select. The free Egg plan covers 70 products and 30 QR tables with no card.
- Open the Plugins directory in ChatGPT, search for “Duckhub QR Menu”, open it and select Connect.
- Sign in with the Duckhub account and approve the consent screen. No API key, no URL to paste.
One detail that surprises multi-site operators: the connection is bound to one venue at a time, and every connected AI app shares that choice. Two venues run from AI need one API key each rather than two OAuth connections.
What OpenAI’s rules say about availability
The directory the app sits in is itself new. OpenAI’s ChatGPT release notes record the change on 9 July 2026, alongside the launch of ChatGPT Work: “We’re also replacing the App Directory with the Plugin Directory.” Plugins are now where connected apps live across ChatGPT web, desktop, Work and Codex.
That consolidation is the reason a listing is worth reporting at all. A single directory spanning Codex and ChatGPT means the same connection an owner makes on a phone is the one a developer reaches from a terminal.
The OpenAI Help Center documentation is also specific about who can reach a given plugin:
“The directory is available across ChatGPT plans, although the label and available features can vary by rollout, surface, and account. Installing or using an individual plugin depends on your plan, workspace, role, region, and its included capabilities.”
We added and ran the app on a standard free ChatGPT account, the plan every user gets by default. The case where that generalisation breaks is a managed Business or Enterprise workspace: there, an administrator decides which plugins staff may install, and OpenAI’s documentation tells users who cannot find a plugin to ask an administrator to review the installation policy.
Why this matters beyond one app
The shift is from recommendation to operation. For three years the restaurant angle on AI assistants was discovery: whether ChatGPT would name your restaurant when somebody asked where to eat. That question is still live, and our restaurant review statistics hub covers how AI recommenders now sit alongside Google in that decision. But it treats the restaurant as a subject the assistant talks about.
Operation is a different relationship. It depends on an open standard: MCP, the Model Context Protocol, which Anthropic released in November 2024 and which has since been adopted across the industry as the way assistants call external tools. Because the standard is shared, the same menu toolset reachable from ChatGPT is reachable from Claude through a custom connector, from Grok the same way, and from Claude Code, Cursor or Codex with an API key. A venue is not picking a side; it is connecting whichever assistant it already uses. Our comparison of which AI assistants can build a QR menu tracks where each one currently stands.
The practical stake for an independent restaurant is time rather than novelty. Menu work is the administrative task most often deferred: the seasonal change that never gets made, the sold-out item still listed at 9pm, the translation nobody has time to do. Those are all small tasks with a login and a dashboard in front of them, which is exactly the shape of work that gets easier when the interface is a sentence.
The honest caveats
Three things this does not change.
AI still cannot photograph your food. It can generate an image, but a generated dish is not the dish that arrives, and guests notice the gap faster than they complain about it.
Prices and allergens still need a human pass. Language models are excellent at structure and merely good at reading numbers out of a photographed menu. A transposed digit costs money; an invented “gluten-free” label is a safety problem. The draft-then-publish flow exists so that review happens before guests see anything.
Adoption of AI tools in restaurants is still early. Industry surveys have consistently put operator use of AI-related tools well below universal, and a directory listing does not change the underlying rate. What it changes is the ceiling: the tools an operator can reach without buying anything new.
For the wider numbers on where AI adoption in restaurants actually sits, see our AI in restaurants statistics hub. For the step-by-step version of the setup above, including the Claude and API-key routes, see can ChatGPT create a restaurant menu?.