Launching in New York

The booking layer for AI agents.

Any assistant can find and reserve a table at a Tadka restaurant through one open protocol: MCP, REST, llms.txt and OpenAPI.

agent ⇄ tadka.xyz/mcp live directory
MCP server
tadka.xyz/mcp

OAuth sign-in as the diner. Search, availability, booking, cancel, waitlist.

REST API
/api/public/v1

Open read-only directory and restaurant detail. CORS enabled.

llms.txt
/llms.txt

Plain-text directory any model can read at crawl time.

OpenAPI 3.1
/openapi.json

Machine-readable spec for function-calling agents.

The problem

Why agents can't book a table today

Bot walls

Reservation sites block automated traffic, so agents fall back to brittle scraping.

Open protocol

A sanctioned, documented interface built for agents to call.

No booking API

Most restaurants have no public endpoint an assistant can transact against.

Free self-serve listing

Restaurants go live in two minutes and become bookable by every agent.

Double-booking

Agents acting in parallel collide on the same table.

Atomic booking engine

Every reservation is checked and written in one transaction against live capacity.

Out of sync

A table booked by an agent never reaches the host stand.

Two-way POS sync

Signed webhooks push every booking; inbound sync holds seats taken elsewhere.

Quickstart

Connect in under a minute

Claude, ChatGPT and MCP clients

Settings → Connectors → Add custom connector, paste the address, sign in and approve.

https://tadka.xyz/mcp

Any agent over HTTP

Read the open directory without authentication.

curl "https://tadka.xyz/api/public/v1/restaurants?city=New%20York"
Full developer reference →

For restaurants

Be bookable by every AI assistant

Free listing

No setup fee, no monthly charge, no commission during launch.

Live in 2 minutes

Add your details, seating times and covers. Self-serve.

Reservations where you work

Partner dashboard with live refresh, or straight into your POS.

List your restaurant