A person tells their AI assistant what local service they need and where. onepick picks one business and books it, even when that business only has a phone.

For Meta's platform team: what the connector does, why local services is an open directory category, the rule behind the pick, and the ten steps you follow on the demo.

Documentation

The pick cannot be bought.

No business can pay to be picked, and no commercial relationship changes the ranking. Every pick discloses any commercial relationship with the business. A runner-up is held in case the first business does not come through. The reasons come only from this list:

  • We measure distance from the requested location.
  • We check state license status where the state publishes it.
  • We count how many independent sources list the business.
  • We check whether it has a website and takes bookings online.
  • Our own record tracks whether it answers the phone and books.

No third-party star ratings, Yelp reviews, or Google Places entries are fetched, stored, or shown. For a person using Muse, your assistant gives a recommendation based on a rule they can read, not a placement someone paid for.

Built to the platform's rules.

Standards

  • MCP over Streamable HTTP at /api/mcp, serving revision 2025-11-25
  • OAuth 2.1 authorization code with PKCE (S256)
  • RFC 8414 and RFC 9728 metadata; an unauthenticated request answers 401 naming the resource metadata
  • client ID metadata documents, with dynamic client registration as the fallback
  • the same four tools over REST under /api/tools, described by an OpenAPI 3.1.0 document at /openapi.json
  • MCP annotations on every tool; every outcome a person can cause is a status, never an error

Consent

  • book is two steps; the first shares nothing and returns a single-use token that lasts 30 minutes, and only the second creates a booking
  • the business hears the person's first name, callback number, what they need, and when they are free, nothing else
  • every AI call says in its first sentence who is calling, that it is an AI assistant calling for a customer, that the call is recorded and transcribed, and gives a callback number
  • AI calls run only 9am to 6pm in the business's local time, only to landline and internet numbers, and never in Indiana, Louisiana, Minnesota, Montana or Nebraska
  • a business stops the calls at once and permanently with one form; nobody can pay to be picked and a booking call is never a sales call

Data

  • delete_my_data removes the account, picks and booking history, and in demo mode the reviewer account stays
  • no payments, deposits or fees in v1, so no card details are held
  • the connector reads nothing from the assistant's memory, chat history or files
  • no Google Places or Yelp content; sources include state license registers and public websites, with every notice on /docs
  • marketing and documentation pages read with JavaScript turned off

After the person confirms, one of three things happens.

  • The business takes bookings online (Booksy, Vagaro, Fresha, Square Appointments or StyleSeat): the person gets its public booking page and no call is placed.
  • An AI assistant calls, announces itself, asks for the service inside the times given, and reports one result: confirmed with a date and time, declined, no answer, voicemail, or the business wants the person to call.
  • No AI call can be placed: the person gets the number and a short script.

Ten steps on the demo.

Use these ten steps to test the connector against our Hoboken test business. The reviewer account is in the notes submitted with the listing, or on request.

  1. Step 1.

    Add the connector: the MCP endpoint is https://app.onepick.smartfront.ai/api/mcp, over Streamable HTTP.

    An unauthenticated request answers 401 with OAuth resource metadata, which tells the assistant where to send you.

  2. Step 2.

    Sign in.

    A person signs in with a one-time code sent by email. The reviewer account is in demo mode: choose "Reviewer? Use a password" and sign in with the password in the notes submitted with the listing.

  3. Step 3.

    Approve the consent screen.

    The assistant holds a bearer token (OAuth 2.1 with PKCE) and lists four tools, each with its annotations.

  4. Step 4.

    Ask: "I need a haircut near 07030 on Friday between 5 and 7pm, pick someone." find_pick

    Status picked: Onepick Test Barbershop, 1 Test Way, Hoboken, NJ 07030, the criterion it was picked on, and the disclosure that the pick is not paid for. The demo account is always given this test business.

  5. Step 5.

    Ask it to book, and give a first name and a callback number. book

    Status needs_confirmation: the business name, the four details that would be shared (first name, callback number, what you need, when you are free) and a single-use token that lasts 30 minutes. Nothing has been shared yet.

  1. Step 6.

    Approve the preview. book

    Status calling, with a booking id: "We are calling Onepick Test Barbershop now." The call is simulated: nothing is dialed.

  2. Step 7.

    Ask for the booking status. booking_status

    Status confirmed, with the appointment time, attempts 1, and the note "Simulated call for the reviewer account. Nothing was dialed."

  3. Step 8.

    Have the assistant repeat the book call with the same token. book

    The same booking comes back. No second call is placed.

  4. Step 9.

    Ask it to delete your data, with confirm set to true. delete_my_data

    The counts of deleted bookings and picks. The demo account stays, so the steps can be repeated.

  5. Step 10.

    Optional: call the same tools as REST, for example POST https://app.onepick.smartfront.ai/api/tools/find_pick with the same bearer token.

    https://app.onepick.smartfront.ai/openapi.json is the OpenAPI 3.1.0 document describing all four.

  • The open category.

    As of 09/20/2026 the Muse directory holds no connector for salons, barbers or trades; only OpenTable covers appointment booking. Muse can book through its own browser without a connector, while onepick books through the business's own booking link or by phone.

    Muse's platform page

  • The ask.

    A conversation with your team, and to be the directory's local-services case study.

    the one-page brief