This MCP change shipped quietly on Sept 5 -- read carefully if you’ve built any automations around the get-reservations tool. Huge thanks to
The change: get-reservations now accepts date_query=booked_at (plus start_date/end_date), a booked_at timestamp filter, and sorting by booking time.
🔍 Our routine: BioScraper
For each new Airbnb booking, BioScraper reads the guest's public profile and the reviews they've left for other hosts, then writes a short summary into the reservation notes. That way we see guest background and any history of harsh reviews before check-in.
Every run starts with one question: which reservations haven't been screened yet?
😩 Before: scan every future stay
The only filter available was by stay date. So we pulled every Airbnb reservation in a 12-month window and checked each one's notes.
On 10 listings, that meant 96 reservations per run to find 5 that needed work, including bookings from as far back as February. It also left a blind spot: anyone booking more than 12 months out was never screened.
😎 After: ask only for what's new
Now we just ask for Airbnb reservations booked in the last 30 days. The window deliberately overlaps the previous run, so there's no state to store. If a run gets skipped, a longer lookback is one word away.
📊 The numbers (measured Sept 11)
- Rows per run: 96 → 39 (−59%), or 7 with a 7-day lookback (−93%)
- Data per run: ~120 KB → ~48 KB (or ~9 KB at 7 days)
- Oldest booking re-read: 6½ months → 30 days
- Blind spot: gone, since stay dates don't matter anymore
- Speed: the same, 1 call and ~2 s either way
Why it matters ⚡
- Tokens, not seconds. If an AI agent is calling Hospitable over MCP, every row costs context. Our old pull was too big for the agent's tool output, so it had to be saved to disk and parsed with a script. Now it comes straight back and gets read directly.
- It doesn't grow with your calendar. The old approach got heavier as bookings piled up further ahead. This one only scales with how many bookings you took in the lookback period.
- Tight windows are tiny. "Everything booked since Sept 8" (3½ days) across all 10 listings and every channel returned 6 rows. That's ideal for anything that polls daily.
🎚️ Picking a lookback
30 days is forgiving: you can skip runs for weeks and still miss no one. 7 days is leaner, but only if the routine reliably runs at least weekly.
💡 Pro tip
Check booking_date on the rows you get back. If anything shows up from well before your window, the filter wasn't applied, and you want to catch that before you process the wrong set.
Your move 👉 If any of your routines find "new" reservations by pulling a stay-date window and filtering on your side, try date_query=booked_at with a lookback, sorted newest first.
Thanks again to the Hospitable team for listening and shipping it! 🏖️