Connect Your Restaurant to ChatGPT and Claude: Manage Your Menu from a Chat

Get started free →

You can now connect your restaurant to ChatGPT or Claude and run day-to-day menu operations from a chat window: check tonight's revenue, 86 a dish, change a price, write a description in three languages — by typing a sentence. Disho ships a remote MCP server that both assistants speak natively, so one connector covers both. No plugins, no code, no developer. Below: the setup, what the AI assistant for restaurant management can actually do (and deliberately can't), and why the security model deserves a second read.

What does it mean to connect your restaurant to an AI assistant?

It means your Disho account becomes something Claude or ChatGPT can act on, not just talk about. Instead of opening the dashboard, you type what you want in plain language and the assistant does it through a secure connection — with your login, your role, your permissions. Real commands that work today:

• "What did we sell best this week?"
• "Put the tartare on the stop-list."
• "Raise the price of the margherita to €12.50."
• "Write an appetizing description for the sea bass in English and Spanish."

The assistant reads your live menu, order queue and analytics, answers in seconds, and makes the change if you asked for one — the difference between a chatbot that knows restaurants in general and an assistant that knows yours right now.

What is MCP, in plain terms?

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to outside tools the way a browser connects to websites — one common language instead of a custom integration per app. Because it's a standard, the same Disho connector at https://disho.net/mcp works with both Claude and ChatGPT: you don't pick a camp, and any future assistant that adopts MCP works too. You never touch the machinery itself — paste one URL, log in, done. That's the whole technical requirement.

How do you connect Disho to Claude?

Three steps, about two minutes, works in Claude on the web and in the desktop apps:

1. Open Claude's Settings → Connectors → Add custom connector.
2. Enter https://disho.net/mcp. Claude discovers the login flow automatically and opens Disho's sign-in and consent page.
3. Log in with your restaurant account and allow access.

That's it — ask "what's on my menu right now?" to confirm. You log in as your own Disho user, so a manager connecting Claude gets manager permissions, not the owner's.

How do you connect Disho to ChatGPT?

Same URL, same account, three steps:

1. In ChatGPT, open Settings → Apps & Connectors and enable Developer mode (a one-time toggle; no development follows).
2. Add an MCP server and enter https://disho.net/mcp.
3. Sign in with your Disho account on the consent page and allow access.

From here on both assistants behave identically — same twelve tools, same protocol, same permissions. Which chat window you prefer is purely a matter of taste.

What can the assistant do — and what can't it?

The connector exposes twelve tools covering menu, orders, analytics, tables and AI content — and deliberately excludes payments, payouts and guest data. What you can do depends on the role of the account that connected:

Three honest caveats. First, every money-spending tool states its price up front, and the assistant is instructed to confirm the cost before generating; failed generations are auto-refunded. Second, deletions are irreversible, and the assistant is instructed to warn you first — but "instructed" is not "physically prevented," so treat delete commands with dashboard-level care. Third, video generation is dashboard-only for now: the chat creates photos and text, but for the 9:16 clips that make a video menu sell, you'll still open Disho itself.

Scenario: the evening "how did we do today?" chat

Ask "how did we do today?" from the sofa and get revenue, average ticket, order count and top dishes in one reply — the same numbers as the dashboard, without opening it. Then keep pulling the thread: "compare that to the last 7 days," "which hour was strongest?", "is the new sea bass actually selling?" The assistant reads today's, 7-day and 30-day analytics and answers each follow-up in seconds.

Scenario: 86-ing a dish from the dining room floor

The kitchen shouts that the tartare is done for the night. Instead of fumbling through dashboard menus mid-service, you open the chat on your phone and type "86 the tartare" — the dish drops off the guest-facing QR menu immediately, so nobody orders it and no waiter has to apologize. Next morning: "bring the tartare back." Staff accounts can't do this one (menu writes are manager-and-up), but a manager on the floor can: one sentence, ten seconds.

Scenario: writing menu descriptions in twelve languages

Type "write an appetizing description for the sea bass in English and Spanish" and the assistant drafts it, you tweak the wording in chat, and it saves the result to the dish — at ≈€0.03 per language. Since Disho menus translate into 12 languages, this is the fastest way to give tourist guests native-quality descriptions instead of machine-translation soup. The AI sticks to the ingredients you give it — important when a guest with an allergy is reading.

Scenario: the five-minute morning routine

Before doors open: "what's on the stop-list? Any orders stuck from last night? How did yesterday close?" Three questions, one coffee, and you know the state of the restaurant before you've put your keys down. If the answer prompts an action — "take the soup off the stop-list, the delivery arrived" — it's the same chat. Prefer it to run itself every morning? The same API powers scheduled workflows — see restaurant automation with n8n.

How secure is giving an AI access to your restaurant?

More controlled than sharing a dashboard password, which is the realistic alternative. The connector uses OAuth 2.1 with mandatory PKCE, and you sign in on Disho's own page, so the assistant never sees your password. Tokens are bound to Disho's server specifically: a token stolen and replayed against another service simply doesn't work. Refresh tokens rotate with theft detection, and only token hashes are stored server-side.

Just as important is what you can see. Every connection appears under Team → Connected apps with its last-activity time, and one click revokes it, killing all its tokens instantly. Permissions follow each user's live role: demote or remove an employee and their connections narrow or die the same moment. And the boundary is structural: nothing related to Stripe, payouts or guest personal data is exposed through the connector at all, so the worst case touches your menu and order queue — never your money. Want to poke at the guest-facing side it controls? The live demo menu is open.

Frequently asked questions

Is it safe to give an AI access to my restaurant? Yes, within clearly drawn limits. Access uses OAuth 2.1 with PKCE, tokens work only against Disho's server, every connection is visible under Team → Connected apps and revocable in one click, and payments, payouts and guest data are structurally excluded. The assistant can only do what the connected user's role allows.

Can the AI spend money without asking? It's built not to: every paid tool (photos ≈€0.15, descriptions ≈€0.03 per language) states its price, and the assistant is instructed to confirm the cost with you before generating. Failed generations are refunded automatically, and only owners can even check the wallet balance.

Which is better for this, Claude or ChatGPT? Honestly — neither. Both connect to the identical MCP server, call the same twelve tools, and inherit the same role-based permissions, so the restaurant-side capabilities are exactly equal. Use whichever assistant you already pay for.

What happens if I remove an employee? Their connections stop working immediately. Permissions are checked against each user's live role on every action, so removing someone kills their access, and demoting them instantly narrows what their connected assistant can do — no hunting down devices or changing passwords.

Do I need a developer to set this up? No. Setup is pasting https://disho.net/mcp into Claude's or ChatGPT's connector settings and logging in with your Disho account — about two minutes, no code anywhere.

→ QR Menu for Restaurants: How It Works and Why Use One