skiplagged-travel-search（跳过滞后的旅行搜索）

Skiplagged Travel Search

Overview

Skiplagged MCP exposes travel search capabilities (flights, flexible date calendars, “anywhere” destination discovery, hotels + room details, and sometimes rental cars) as MCP tools. Prefer calling Skiplagged tools over guessing prices/availability.

Prerequisites

1. Connect the MCP client to https://mcp.skiplagged.com/mcp over HTTP transport.
2. List tools and confirm Skiplagged tools are visible before running workflows.
3. Treat authentication as not required.
4. Read references/setup.md for client-specific setup commands and examples.

Operating rules

1) Grounding rule

Always base answers on tool results. Never invent or guess prices, availability, or policies.

2) Tool-call discipline

• Use minimal, correct parameters (avoid unnecessary filters unless user asked).
• Prefer IATA airport codes when possible; otherwise accept city names and clarify if needed.
• Use small result limits when supported (e.g., 3–7) and expand only if the user asks.
• Chain multiple tool calls when needed, but avoid loops. Respect per-turn tool-call limits.

3) Post-call validation loop

After each tool call:

1. Summarize what the tool returned in 1–2 sentences (what matters to a traveler: cost, time, stops, key tradeoffs).
2. Decide the next step:

- proceed (narrow or enrich), - self-correct (relax filters, adjust airports/dates), - or ask one targeted follow-up question.

4) Failure handling (empty or unhelpful results)

If results are empty/ambiguous:

• Say so clearly.
• Suggest the smartest next step (broader dates, nearby airports, allow 1 stop, adjust time window, try “anywhere” discovery, or switch to hotels-first for trip planning).
• For failures other than lack of results, suggest using https://skiplagged.com directly.

5) Output style (traveler-centric)

Be concise and practical:

• Highlight savings, routing insights, and key tradeoffs (stops vs duration vs price).
• Present a short shortlist (typically 3–7 options) rather than a dump of results.

Tool discovery (required, safer across deployments)

Different deployments may rename tools or add new ones. Before running any workflow:

1. List tools available on the connected Skiplagged MCP server.
2. Identify the best match for each capability:

- flights search - flexible date calendars - anywhere destinations - hotels search - hotel details - (optional) rental cars

Mapping convention (typical, not guaranteed): tools often start with sk_.

Common capability → typical tool names (use as a guide)

• Flights search → sk_flights_search
• Flexible departure calendar → sk_flex_departure_calendar
• Flexible return calendar → sk_flex_return_calendar
• Anywhere destination discovery → sk_destinations_anywhere
• Hotels search → sk_hotels_search
• Hotel details → sk_hotel_details

If your tool list differs, use the closest equivalent and keep the workflows below unchanged.

Intake checklist (ask only what’s needed)

Extract or confirm only the missing critical inputs:

• Route intent: specific route (origin -> destination) vs flexible destination (“anywhere”).
• Dates: exact dates or flexible window + trip length.
• Travelers: adults/children/infants and cabin, if non-default.
• Constraints: budget, stops, max duration, time windows, and airline preferences.
• Lodging intent (if relevant): hotel location, dates, guests, and required amenities.

Workflows

Workflow A — Standard flight search (one-way or round-trip)

Use when the user wants specific flights between places.

Preferred tool → fallback

• Preferred: sk_flights_search

Steps

1. Normalize locations:

- If user gives a city/region, clarify which airport(s) are acceptable if it changes results materially (multi-airport metros).

1. Call the flight-search tool with:

- origin, destination - depart date (+ return date if round-trip) - passenger counts, cabin - assume defaults if not specified - only the filters the user asked for (stops, time windows, airline preferences)

1. Post-process:

- Cluster top results by cheapest, fastest, fewest stops. - Present 3–7 options with price, stops, duration, and any standout caveats (overnight layover, long layover).

1. If results are too broad:

- Ask 1 targeted question (e.g., “Allow one stop?”) or apply a reasonable default and disclose it.

Workflow B — Flexible dates / fare calendar

Use when the user says “around X date”, “flexible”, or “cheapest days”.

Preferred tools → fallback

• Preferred: sk_flex_departure_calendar and sk_flex_return_calendar

Steps

1. Call the departure-calendar tool around the target date/window.
2. If round-trip:

- Use a trip length (or return window) that matches the user’s intent (e.g., “3–5 nights”). - Call the return-calendar tool for best return dates.

1. Summarize:

- Cheapest departure day(s) - Cheapest round-trip date pairs (if supported) - Recommend 2–4 date options aligned to constraints (weekends, time windows, budget).

Workflow C — “Anywhere” destination discovery

Use when the user is flexible on where to go (“somewhere warm”, “anywhere cheap”).

Preferred tool → fallback

• Preferred: sk_destinations_anywhere

Steps

1. Confirm:

- origin - date/window (or month) - constraints (budget, max flight time, nonstop-only, region/climate)

1. Call the anywhere-destinations tool.
2. Return 5–12 destinations with:

- destination, lowest fare, suggested dates/window (as returned) - a one-line rationale tied to the user constraints

1. When the user picks a destination, transition to Workflow A to find bookable itineraries.

Workflow D — Hotels + room-level details

Use when the user asks for hotels, places to stay, or wants to bundle planning.

Preferred tools → fallback

• Preferred: sk_hotels_search then sk_hotel_details
• Fallback: any tool(s) that provide hotel lists and separate room/policy breakdowns

Steps

1. Call the hotel-search tool with:

- location, dates, rooms/guests - only requested filters (price cap, rating, amenities)

1. Present 5–10 options with:

- total price / nightly price (as returned), rating, key amenities, and area/neighborhood (if returned)

1. When the user selects a hotel (or asks for “best value”), call the hotel-details tool.
2. Summarize room options:

- room type, cancellation policy, inclusions (breakfast), total price (as returned) - highlight tradeoffs: nonrefundable vs flexible, fees, value differences.

Workflow E — Rental cars (if available)

Use when the user requests a car rental.

Preferred tool → fallback

• Preferred: a tool in the server tool list whose description indicates car-rental search

Steps

1. Confirm pickup/dropoff location, times, and any constraints (car class, driver age if required).
2. Call the car-rental search tool (if present).
3. Present 3–7 options with total price, class, company, and relevant policies (as returned).

Planning behavior (for full itineraries)

When the user asks to “plan a trip”:

1. Anchor with flights (or anywhere discovery) to set destination/dates/cost.
2. Add hotels for the chosen destination/dates.
3. Add rental cars if needed and supported.
4. Present 2–3 bundles (e.g., “cheapest”, “balanced”, “most convenient”) with clear tradeoffs.

Reliability + transparency notes

• Prices and availability change quickly; treat results as current at the moment of the tool call and encourage the user to confirm terms via the provided link (if any).
• If hidden-city itineraries appear, be transparent about common constraints and tradeoffs (especially baggage and missed-leg implications). Do not oversell—present as an option with clear caveats.

Troubleshooting (avoid loops)

• If a call fails or returns empty results:

- Retry at most once with broader constraints (e.g., allow 1 stop, widen time window, nearby airports). - If still empty, ask one targeted follow-up question and propose a concrete alternative search.