PMS action integration
The wiring that lets an agent actually touch the property-management system: look up a guest or a bill, write back a task, and know exactly where the wall is.
A dashboard can only read. For an agent to do real front-office work it has to reach into the property-management system itself, to look up a guest, pull a rate, find a reservation or a bill, or write back a task or an order. I wanted a scoped, safe way to hand Claude exactly those actions without handing over the whole PMS.
How it works
The integration runs through the Zapier MCP, scoped to a single Mews account. It exposes three kinds of action.
- Five keyed reads: get a customer, a rate by ID, a reservation by customer or by reservation ID, and a bill by ID.
- About a dozen writes: create or update a reservation, customer, company, order, task, accounting item, or availability block.
- Trigger-type events for reservation, customer, and closed-bill changes.
The valuable thing I learned is the boundary. Every read is keyed by a specific ID. There is no list-reservations-by-status-or-date action, so you cannot run a retrospective scan like find every cancellation this week through this connection. The only thing that can filter on a canceled state is a live event trigger, and that fires only inside an always-on Zap, never on demand.
So this is the fast path for keyed, event-driven work, and it is live. For the bulk and historical reads it cannot do, I am building a vendor-neutral data layer alongside it, pulling the Mews Connector API getAll endpoints into a local SQLite spine. The two are complementary: Zapier for cheap keyed actions and triggers, the Connector API for the full-history reads. Together they are the wiring that lets one person plus an agent act on the PMS the way a front desk used to.