What you can evaluate today.
Start with place search and booking requirements in the San Francisco pilot. Separate a useful read, a submitted request and a confirmed reservation.
Search and place details
Search and place details are the pilot’s read-only starting point. Check the actual returned fields, location and missing information with your authorized account.
Try the dinner shortlistBooking
Native WhatsDo exposes tools for booking requests and follow-up. Requirements vary by venue, and some requests need manual follow-through. Keep a request pending until the supplying system confirms the intended details.
The local toolkit prepares a brief and stops before sending. It exposes no booking or payment action.
Review the booking flowPayments and other fulfillment
External agent payments are not available in this pilot. The local trial does not order coffee, buy tickets or complete hotel reservations. Those outcomes require a separate capability review.
MCP and API access
The MCP endpoint requires authorization. OpenAPI and OAuth discovery responded publicly on 2 October 2026; anonymous MCP requests received an authentication challenge. That check does not establish an authenticated production session.
Connect with your own accountChoose the right evaluation package
The new agent toolkit runs three read-only scenarios and an MCP App using synthetic samples or explicitly authorized live reads. The separate REST evaluation kit remains a narrower search runner with its own credential model.
Open the agent toolkit Inspect the REST evaluation kit