Skip to main content
Documentation
Service VIN developer & AI platformGetting startedAuthenticationPaginationIdempotencyRate limitsErrorsWebhooks & signature verificationConnecting ClaudeRecipesMCP servers & the Claude connectorREST resource referenceCapability indexCustomers capabilitiesVehicles capabilitiesLeads capabilitiesWarranties capabilitiesQuotes capabilitiesInvoices capabilitiesPayments capabilitiesServices capabilitiesJobs capabilitiesCalendar capabilitiesScheduling capabilitiesStaff capabilitiesInventory capabilitiesComms capabilitiesTasks capabilitiesShop capabilitiesDocuments capabilitiesKnowledge capabilitiesReports capabilitiesBooking capabilitiesForms capabilitiesMarketing capabilitiesReviews capabilitiesAutomations capabilitiesIntegrations capabilitiesChangelogSearch

Guide

Pagination

Two paging models, on purpose: an opaque keyset cursor on the REST resources, and an exact remaining count on the capability lists.

The REST resource routes page by keyset cursor. Read meta.next_cursor from a page and send it back as ?cursor= to get the next one; when it comes back null, you have reached the end.

  • The cursor is opaque. It encodes a composite position, not a timestamp. Do not parse it, build one, or compare two — just hand it back.
  • Page size is 25 by default, 100 at most. Out-of-range and unparseable values are clamped or fall back rather than erroring.
  • New rows never shift your page. Paging walks backwards from a fixed point, so an import landing mid-sync cannot make you skip or repeat a record.
  • Webhook subscriptions are not paginated — that list is capped at 10 per shop and returns no meta block.

Capability lists page differently, and that is deliberate

A capability that can return a partial answer declares a truncation contract, and its response carries an exact count of what it did NOT return — omitted, with 0 as a real answer meaning you have all of them — plus the way to get the rest.

Silent truncation is the most dangerous thing a read can do to an agent

A person who gets 25 of 300 customers sees a page control and knows. A model gets an array, has no page control to look at, and answers the owner's question confidently and wrongly about a book it has seen a twelfth of. So a list capability may not return a bare capped array: it says how much it withheld, every time, and its operating contract tells the model what to do about it.

A cursor is right for a machine walking an entire table in order, which is what a nightly sync does. An exact remaining count is right for an agent asking a narrow question and then answering a shop owner out loud. Both surfaces read the same rows through the same mapper, so the records are identical either way.