Repository navigation
Local services and appointment booking for small merchants: requirements and an offer to help with EP #317 #1022
Replies: 4 comments 7 replies
|
Transitioning from traditional cart-based physical commerce to appointment scheduling is fundamentally an inventory-modeling challenge. Physical goods consume discrete units ( To keep the protocol clean without overloading the core order pipeline, a few structural considerations from scheduling architectures:
Great initiative bringing real-world SMB service constraints into the specification! |
|
Your distinction between a request and merchant confirmation also matters for repairs and cleaning, where scope and duration may not be known at first contact. I operate Agent Service Exchange. We have an active network of at least five service providers covering plumbing/heating, electrical work, air conditioning and cleaning; the current public MCP pilot supports plumbing/heating and cleaning enquiries in Vienna. We require approval for the exact recipient and shared data, then return an enquiry receipt with For your One reproducible test case is an agent retrying the same approved enquiry: it must return the same receipt without creating a second request. We have a runnable synthetic MCP example for that boundary, with email disabled: Our current implementation uses MCP; UCP compatibility has not been implemented or validated. |
|
Quick status update, since my first post is now out of date. When I opened this I wrote that we had no MCP server. We do now, and it is in production: one MCP endpoint per business, enabled only when the owner opts in. It exposes four tools:
Two rules from this thread are enforced there today: nothing is confirmed without the merchant (the pending_merchant_action idea), and nothing is sent to the phone number an agent supplies until the end customer proves they own it. Built and going out in our next release, not live yet: setup/teardown buffers, concurrent services, a machine-readable quote acceptance step (the enquiry → quote_offered → quote_accepted flow Patrick's question led to), and idempotent retries of request_booking. Still no UCP endpoint. Next from our side: a field-by-field mapping of this flow onto the #317 schemas, listing every place a small local merchant does not fit, plus JSON Schema files for the extension. One question for the maintainers: #317 has been quiet since August. Is it still the right place for services work, or is there a newer thread or working group we should follow? |
|
Hi all, Timur from Meta (the platform on the #922 charter). I work on catalog support for service businesses, so this is catalog-side input to sit alongside the scheduling model in this thread. Taking the USP Service schema that UCP-native mode would inherit as the base, some gaps stand out from what we have needed to make services discoverable and routable at scale. 1. A structured service area for at_buyer_location
2. A quote and enquiry path, not only bookable slotsMany service businesses cannot give a price or a duration until they understand the job (renovation, repairs, legal or financial advice, consulting). They convert through an enquiry, a message thread or a call rather than a slot.
Related: keep the minimum service record small. Most small service businesses have never maintained a structured catalog: their offer lives on a website or a business profile and gets structured later. If a fixed price, an exact duration and full policies are effectively required, those businesses stay invisible to agents. An 3. Provider credentialsIn regulated or trust-sensitive categories (health and beauty treatments, gas and electrical work, legal and financial advice), a licence or certification is often what the buyer decides on, and what a platform has to check before showing the service. I could not find a field for them in the Service or Provider schema. A sketch, on Platforms should treat these as asserted by the business unless they verify them. Where a register or an accredited body signs a credential, it could travel as one of the verifiable attestations proposed in #1020. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Hi all. I run CorujaAí, a small SaaS in São Paulo, Brazil, for very small local businesses: pet grooming, hair and nail salons, beauty studios, small clinics, and small tour/experience operators (e.g. a boat tour with fixed departures and limited seats). I'd like to add requirements from that segment to the appointment scheduling work in #317 and offer to build a reference implementation. I'm new to this community, so please correct anything I'm getting wrong about scope or process.
The problem
Today UCP models a merchant selling items: Catalog, Cart, Checkout, Order. A dog grooming appointment isn't an item in a cart. What the customer buys is:
When the buyer is a personal agent acting for an end customer, there is one more piece: consent. The merchant receives personal data (name, phone, sometimes a pet's or patient's details) from someone it never talked to, and it needs to know the end customer authorised this booking.
The lodging draft (
dev.ucp.lodging.booking) already covers several of these ideas for hotels: real-time availability, deposits and deferred balances, cancellation and refund policies tied to the booking lifecycle. #317 proposes a general services flow (SearchAvailability,HoldSlot,BookAppointmentwith a TTL hold and apayment_reference). I think most of what small local merchants need fits on top of #317. The points below are what I'd add.Sketch of capabilities
Names are placeholders; I'm happy to follow whatever naming #317 and the TC settle on.
scheduleidea in Use Cases Outside Retail #93/thinking about services #163).pending_merchant_confirmationwith a deadline, thenconfirmedordeclined. Many small merchants can't accept fully automatic bookings, and the agent needs to tell the end customer "requested, not yet confirmed".Why Brazil is a useful test case
What we have and what we don't
What CorujaAí already has in its codebase: public availability and booking endpoints per business, customer self-cancel and reschedule, deposits by Pix to the merchant's own key, and a WhatsApp AI assistant for the merchant with an "owner approves before it's sent" mode.
What we don't have yet: an MCP server or a UCP endpoint. Nothing here is a working UCP implementation today.
Offer
If it helps #317, I'd like to:
Questions for the group:
Related: #317 (Enhancement Proposal: Universal Appointment Scheduling), #163, #93, #864 (Lodging booking draft)
Thanks for reading. Happy to adjust or fold this into #317 directly.
cc @ejoslin — would love your thoughts on whether these small-merchant requirements fit #317
All reactions