HotelRefund Price Tracker MCP Server

Show what one hotel (agoda_hotel_id from search_hotels) costs on Agoda for each check-in date in a window — the "what if I move my dates" view. Returns one price snapshot per date (with fetched_at), the cheapest date, and low/high bands for the hotel; dates with an active tracking are marked is_tracked with their track_page_url. This is a cross-section by date, not a time series: to keep watching a date, call track_hotel_price for it. Most dates are fetched in the background on the first call, so pending may be > 0 — call again with the same arguments after ~3 seconds to get the rest. Default window: 28 days from start_date (or today), 1 night, 2 adults.

List today's biggest observed hotel discounts: Agoda's live rate for a 1-night stay about two weeks ahead (2 adults, USD) compared with Agoda's own listed rate for the same room; only discounts of 5% or more are kept. Without arguments, returns the single best deal per city across major cities worldwide, biggest discount first. With city (a city name in Korean, Japanese, Chinese or English) or city_id, returns the current deals for that city (up to 6; an uncached city is fetched live at most once per 6 hours). Each deal includes booking_url, the Agoda page for that hotel and stay, and image_url when a photo is available.

Resolve a hotel to the agoda_hotel_id that track_hotel_price needs. This is a lookup step, not a hotel search engine: it returns catalog entries (name, address, stars, rating, review count) and no prices — for prices, track the hotel or use list_hotel_deals. Give a city, region or airport name (Korean, Japanese, Chinese or English) and, when the user named a hotel, the hotel filter; the city is resolved to the best catalog match. If several cities match, other_city_matches lists them — pass one of their city_id values (instead of city) to search that city. Typical flow: search_hotels → track_hotel_price → get_price_history.

Return the recorded prices for one tracking (tracking_id): current, first, lowest and highest price, percentage change and trend since tracking started, and the time series of observed prices. Also returns booking_guidance — a book-now-or-wait signal computed from this stay's own observations, with the trend slope, volatility, days left, a short-horizon predicted range, and headline + summary sentences (in language: en, ko, ja or zh) that explain the read the same way the page does, including when a price jump was ignored as a level shift. track_page_url is the booking-decision page in that language. And report_url / csv_url: every observation with UTC timestamps as a printable page and a CSV, which people use as evidence for best-price-guarantee claims, refund requests and check-in negotiation. When available, seasonality adds the destination's forecast monthly rates for the hotel's star band (a separate scale from the tracked prices — it says whether the stay month is a high or low month for the city, and which nearby month is cheapest). A tracking created moments ago may report awaiting_first_price until its first check completes.

Return the full text of one HotelRefund reference guide on what to do about a booking that already exists: cancellation deadlines, getting a refund, what to do when the price drops after booking, how best-price guarantees work, refundable vs non-refundable rates, and why rates move. Call without arguments to list every guide with its topic id and a one-line summary. Guide text is English; language returns the matching Korean, Japanese or Chinese page url alongside it.

Report whether the tracker is running and what it covers: the number of hotels and cities in the catalog that can be tracked, how many stays are tracked right now, price checks in the last 24 hours, and the remaining shared daily registration allowance for AI clients. Useful before starting a longer flow, or to confirm the server is live.

Today is 2026-09-10 (UTC) — never infer the date from the conversation or from training data. Stay dates are calendar days in the hotel's local time zone, which in Asia is already ahead of UTC, so check_in must be 2026-09-09 or later; an earlier check_in is rejected, not silently accepted. Start tracking one hotel (agoda_hotel_id from search_hotels) for one stay (check_in to check_out, YYYY-MM-DD). This is the tool that creates lasting state: from now on the Agoda price is checked automatically several times a day and every observed price is recorded, so the same tracking_id can be asked about days or weeks later with get_price_history. Free, anonymous and public; no account. If the same hotel and dates are already tracked, the existing tracking is returned instead of a duplicate, so it is safe to call again. The response includes tracking_id, booking_url (the Agoda page for this hotel via a HotelRefund redirect) and track_page_url (the booking-decision page: price chart, timing read and a book button; in the language given by language, with track_page_urls for all four). Used both before booking (to catch a drop) and after booking a refundable rate (to see when rebooking would pay off).

List the hotels currently being tracked (all trackings are public), with current price, change since tracking started, stay dates and per-item booking_url / track_page_url (in language), sorted with the biggest price drops first.