MCP: Travel Fare & Hotel Price Intel

An MCP (Model Context Protocol) server wrapper around travel-fare-intel, exposing structured flight-fare and hotel-price data as AI-agent tools. Because Google Flights has no public API (and ITA's was discontinued), this reads the open, logged-out result pages and returns clean, normalized fares an agent can reason over.

Tools

Tool	What it does
search_flight_fares	Current fares for one route (origin/destination/date, optional return), cheapest-first: carrier, USD fare, cabin, stops, duration, times, booking source.
get_route_grid	Cheapest fare per departure date across a window for one route - the date grid for "which day is cheapest?".
watch_fare	Fare-watch one route: current cheapest fares plus price drops vs the previous run (pair with a Schedule for recurring alerts).
search_hotel_prices	Hotel prices for a destination + check-in/check-out window, cheapest-per-night-first: total + per-night price, stars, review score, room type.

Modes

• list_tools - emit the tool catalog (name, description, JSON schema per tool) plus the agentic-payment descriptor. Free.
• call_tool - invoke one tool: set tool and args.
• batch - invoke several {tool, args} calls in one run (max 10).

{ "mode": "call_tool", "tool": "search_flight_fares",
  "args": { "origin": "JFK", "destination": "LAX" } }

Dates and validation

• Every date argument is optional and rolls relative to today (UTC): departure_date defaults to 30 days out, get_route_grid defaults to the 30-day window starting two weeks out, and search_hotel_prices defaults to a 2-night stay 30 days out. The dates actually used are echoed back in args.
• Arguments are checked before any charge or upstream run: 3-letter IATA codes, YYYY-MM-DD dates, no past dates, nothing more than 330 days out, return after departure, grid windows up to 60 days, hotel stays up to 30 nights. A bad argument returns ok: false with a readable error, and the run still finishes SUCCEEDED.
• If the upstream actor answers with sample output instead of live results, or finishes without records, the tool returns ok: false with the upstream's reason rather than passing sample data off as live fares.

Agentic payments (x402 + Skyfire)

This server is x402 (USDC on Base) and Skyfire ready. When the operator enables Apify's MCP monetization, an AI agent can pay per tool call with no pre-provisioned API token. The list_tools response advertises the accepted rails in its payments block. Enable via env: X402_ENABLED, X402_PAY_TO_ADDRESS, SKYFIRE_ENABLED, SKYFIRE_SELLER_ID.

Connect it to an MCP client

This Actor is exposed as a remote MCP tool through Apify's hosted MCP server. Point any MCP-compatible client (Claude Desktop, Cursor, VS Code, or an OpenAI / LangChain / LlamaIndex agent) at Apify's server with this Actor enabled:

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com?tools=seibs.co/mcp-travel-fare-intel",
      "headers": { "Authorization": "Bearer <YOUR_APIFY_TOKEN>" }
    }
  }
}

Get a token (free) from Apify Console -> Settings -> API & Integrations. The Actor then appears as a callable tool. Run mode=list_tools first (free) to fetch every tool's live JSON schema, then call one:

{ "mode": "call_tool", "tool": "search_flight_fares", "args": { } }

Prefer a direct call? Hit the Actor straight through the Apify API / SDK - no MCP client required:

from apify_client import ApifyClient
client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("seibs.co/mcp-travel-fare-intel").call(run_input={
    "mode": "call_tool", "tool": "search_flight_fares", "args": {},
})
items = client.dataset(run["defaultDatasetId"]).list_items().items

Pricing

Flat $0.005 per MCP tool call plus upstream pass-through (the travel-fare-intel actor charges its own fare_record / hotel_record / route_history / price_drop_alert events to the same run). A hard cap of 5 tool calls per run, a $0.05 minimum-revenue abuse guard, and a 600s runtime cap protect against runaway agent loops.