Restaurant menus, tables and delivery answered by the guest's AI
How a restaurant or a group answers diners with each guest's own AI, which MCP tools it serves, and which interface suits its site.
Published on . Updated on
In a March 2026 survey of 3,001 US consumers, 22% had used an AI tool to choose a restaurant and 75% were comfortable using AI for reservations. Of restaurant operators, 39% have updated their listings to appear in AI answers. AI Connect, the TensorPass SDK a website installs, puts that answer on the restaurant's own site. The guest's model reads the menu or the bookings through the site's tools and answers on the guest's own account through AI Wallet, so the site pays no inference.
Where diners meet AI in 2026
Each service completes the booking or the order in the partner's own system:
- By August 2026 ChatGPT took table bookings through Yelp in the US and Canada, through OpenTable and through Resy.
- DoorDash runs a grocery app in ChatGPT, and its checkout runs in DoorDash's own app.
- Uber Eats in ChatGPT shows restaurants and menus in the US, and the order goes through the Uber Eats app.
A restaurant site with AI Connect answers on its own pages the same questions guests put to those apps:
- For "Which of your restaurants is open late tonight?", the model calls
list_locations. - "Is the risotto gluten free?" calls
get_menuwith a diet filter. - "Do you have a table at 8 pm on Friday?" calls
check_availability. - "Do you deliver to my street, and until when?" calls
get_locationfor the delivery zones and hours.
Tools for one restaurant or a group
The Model Context Protocol (MCP) is the open standard through which a model calls a site's tools. A single restaurant or a group serves these tools:
| Tool | Input | What it returns |
|---|---|---|
get_menu | Section, diet | The dishes with their price, allergens and diet labels |
check_availability | Date, time, party size | The free slots, and the address of the booking page for each |
get_location | Location id | Address, opening hours, phone, delivery zones and the ordering page |
list_locations | None | The group's restaurants with their city and page address |
The MCP guide sets the rules for naming, describing and bounding each tool.
Which interface suits a restaurant site?
A group with many restaurants suits the side panel (AiChatPanel), which keeps the conversation while the visitor opens a restaurant's page. A single restaurant puts the popup (AiChatPopup) on every page, and an inline chat (AiChatInline) on its menu page answers allergen questions next to the dishes. The full list is under interface modules.
Bookings and orders after the answer
Every tool above reads and changes nothing. A restaurant's check_availability returns the address of its booking page as a field of the result, and its reservation system confirms the table. Delivery orders go through the restaurant's ordering page.
The SDK's answer guard lets a link through when it points to the page's own origin, to a host the site lists, or to an address a tool returned as a field. A booking link returned as a field of check_availability shows as a link. A link written inside the text of a review is no field, and the guard keeps it unclickable.
Next steps for a restaurant site
- WordPress: the WordPress quickstart adds the chat as a block or a shortcode. The plugin is not in the WordPress plugin directory yet and builds from the AI Connect repository.
- Own stack: the React or JavaScript quickstart, then the MCP guide.
- Done for you: the AI native service audits the site, serves its menu or booking as an MCP server, adds an AI mode in the site's design system and hands over the code, in under 24 hours. The result is an AI native website.
On 16 November 2026 the SDK packages of version 0.1.0 are on npm, and AI Wallet waits on the Chrome Web Store's review while a visitor can install it by hand from https://tensorpass.app/wallet. A guest without AI Wallet gets the AI mode through a sign in with OpenRouter, or with the TensorPass plan on a site that passes a site key.
Sources
- DoorDash, Restaurant industry trends report 2026, 2026
- Android Authority, ChatGPT restaurant reservations and waitlists, 10 August 2026
- Digital Commerce 360, DoorDash app in OpenAI's ChatGPT, 19 December 2025
- Uber, ChatGPT integration for eaters, undated
- TensorPass, Expose your tools over MCP, 12 November 2026
- TensorPass, Side panel, 12 November 2026
- TensorPass, Guards, 12 November 2026
Questions
- Can the AI mode book a table?
- The tools on this page only read data. The check_availability tool returns the address of the booking page, and the guest books there in the restaurant's own system.
- How many results should a search tool return?
- Return a short list, such as 10 results by default and 20 at most, best rated first. A short list keeps the conversation small on the guest's plan.
- Who pays for the answers on a restaurant site?
- Each answer runs on the guest's own ChatGPT, Claude or Gemini account through AI Wallet and counts against that account's plan or credits. The site pays for its MCP server and its hosting.