Sign inTry it

OpenTable MCP server: let your AI agent find restaurants by date and party size

Wield is an MCP server that gives your AI agent one OpenTable tool. The tool searches restaurants in a city by cuisine, name or vibe, for a date, time and party size. It runs in your own browser with your own session and needs no login. In our benchmark the recipe passed 9 of 9 runs.

Updated

Live recipeopentable.com/search-restaurantsv1

OpenTable Find restaurants to book on OpenTable for a date, time and party size

Read-only, no loginRuns in your browser only

Parameters (5)
citystringCity, e.g. 'New York', 'San Francisco', 'London'. Omit to use OpenTable's guess of the user's area
datestringrequired
party_sizeinteger
querystringrequiredCuisine, restaurant name or keyword, e.g. 'italian', 'sushi', 'rooftop'
timestring
Use it with your agent
Measured on the live site
MetricWield recipeBrowser Use + Claude Sonnet 5.5
Success100%0%
Median time1.9 s28 s
Cost per run$0$0.077

9 runs each, 7 November 2026. How we measure

What your agent gets

RecipeWhat it doesLoginStatus (7 Nov 2026)
opentable.com/search-restaurantsFinds restaurants to book: name, cuisine, price band, neighbourhood, rating, reviews, how often it was booked recently, address, maximum party size and the restaurant pageNoOK. 9 of 9 benchmark runs passed

The parameters are query (required, such as a cuisine or a restaurant name), city, date (required), time (default 19:00) and party_size (default 2). In our in-page check, the query "sushi new york" returned 156 restaurants for New York City.

Time slots are not in the output yet. OpenTable loads them separately, through GraphQL, so Wield cannot show free times. Open the restaurant page to pick a time. The how-to page has example prompts.

How a request flows

  1. You ask your agent: "Find a sushi place in New York for four people this Friday at 19:00."
  2. The agent calls find_recipe, then run with opentable.com/search-restaurants and your query, city, date, time and party size.
  3. wield-mcp passes the run through a local bridge to the Wield extension.
  4. The extension sends the request from an opentable.com tab, so it carries your browser session.
  5. The results go back to your agent. Nothing passes through wield.run.

What leaves your computer

A search makes one request to www.opentable.com. Passwords and cookies never leave your computer. Wield logs the request on your computer only, without bodies or query values. Reservations are made by you on OpenTable, not by Wield.

OpenTable terms, in plain words

OpenTable's terms prohibit automated access. The recipe makes one request per search, one search at a time, and you start each one from your own browser. Reservations stay with you on OpenTable. We have not completed a review of the terms for this recipe, so read them before you automate anything else.

Limits and what is not working yet

  • No free times yet. The recipe returns restaurants, not open time slots.
  • Bot checks stop the run. When a site shows a check, Wield stops and asks you to solve it in your own tab. Wield does not solve bot checks. In our runs, the cloud browser agent hit bot protection on all nine OpenTable runs.
  • One search at a time. The recipe is meant for a search you start yourself, not for polling or bulk collection.

FAQ

Is there an official OpenTable MCP server?

Not that we know of as of 7 November 2026. Wield provides this search as a recipe that runs in your browser.

Which AI apps work with it?

Any MCP client. npx -y wield-mcp init sets up Claude Desktop, Claude Code, Cursor and Codex. The AI apps page lists each one.

Does it need a login?

No. The search runs without an account. Booking is done by you on OpenTable.

How fast is it compared with a browser agent?

In our benchmark, as of 7 Nov 2026 (preliminary, still running), the recipe passed 9 of 9 runs with a median of 1.9 seconds. Browser Use driving Claude Sonnet 5.5 passed 0 of 9 runs, with a median of 27.6 seconds and about $0.08 per run. Bot protection blocked the browser agent in all nine runs. The benchmark page has the method and raw logs.