The API surface area: what's exposed, what isn't

Every major PMS, channel manager, and revenue management system has an API. Every operator I have spoken to has assumed this means agentic integration is straightforward. It is not. The gap between "the vendor has an API" and "the API can do what your agent needs" is where 60% of agentic projects stall — and the stall is rarely a vendor bug. It is a mismatch between the actions your agent needs to take and the actions the API actually exposes.

What is usually exposed

The common read-side surface area across Opera Cloud, Mews, Cloudbeds, Apaleo, and Protel is reasonably consistent: list reservations by date range, get a specific reservation by ID, list available room types and rates for a date range, read guest profiles, read folio balances. These are the operations a property management dashboard or a typical OTA channel manager already needs, so vendors have invested in them.

The common write-side surface is narrower: create a reservation, modify dates or room type on an existing reservation, cancel a reservation, post a charge to a folio, post a payment to a folio. These exist on every modern PMS but with varying granularity — Mews and Apaleo expose finer-grained operations than Opera Cloud, which often requires composite operations that can fail partway through.

What is usually NOT exposed (and what to do about it)

Rate limits and throughput

Every PMS API has rate limits that are not documented prominently. Opera Cloud OPI typically allows 5-10 requests per second per property; Mews Connector API is more generous at 50 RPS but throttles aggressively on burst; Apaleo is openly published at 60 RPS. For an agent running 600 sessions per month, none of these are constraints. For an agent running across 40 properties at peak hours, all of them become constraints. Test the throughput you need before you architect the workflow, not after.

The two questions to ask the PMS vendor

Before committing to an agentic project, the operator should get two specific answers from the PMS vendor in writing. First: "Can I read the current state of [object] via API, including [specific fields]?" Second: "Can I write [specific operation] via API, with what preconditions, and with what error handling?" Vague answers ("yes, our API supports that") are red flags. Specific answers with field names and error codes are green flags. The difference between the two predicts whether your project ships in 8 weeks or 8 months.