CHEAP_FLIGHTS_MCP

MCP server for searching cheap flights in real-time via RyanAir, WizzAir, and Google Flights.

Built with FastMCP (Anthropic's Model Context Protocol), featuring parallel airline queries via asyncio.gather, exponential backoff retry, and verified booking deep links.

Features

• Real-time prices directly from airline systems (not cached/aggregated)
• Parallel search across multiple airlines simultaneously
• FastMCP - modern declarative MCP pattern (2025 standard)
• Web UI - browser-based interface with dark theme
• AI Agent - conversational flight search via Claude (SAP Hyperspace)
• Deep booking links - click to buy directly on airline website
• Email results - send search results via email (mailto: protocol)
• Configurable airlines - add/remove airlines dynamically
• USD/EUR conversion - live exchange rate for Google Flights results
• Retry mechanism - exponential backoff + jitter for rate-limited APIs
• Security - input validation, generic errors, no credentials in code

Supported Airlines

Airline	Data Source	Destinations	Reliability
RyanAir	ryanair-py library	Routes API (all, incl. seasonal)	Excellent
WizzAir	Unofficial JSON API	Route map	Good (429 retry)
Google Flights	fast-flights library	N/A	Good (verification source)

Installation

git clone https://github.com/IngRobertFodor/CHEAP_FLIGHTS_MCP.git
cd CHEAP_FLIGHTS_MCP
pip install -r requirements.txt

Or via Smithery:

npx @smithery/cli install @ingrobertfodor/CHEAP_FLIGHTS_MCP

Quick Start

Option A: Web UI (no AI required)

python web/app.py

Open browser: http://localhost:5000

Option B: CLI AI Agent (requires AI Proxy)

python agent/flight_agent.py

Option C: MCP Server only (for integration)

python mcp_server/server.py

Configuration

Create a .env file:

AI_PROXY_BASE_URL=http://localhost:6655
AI_PROXY_API_KEY=your-api-key-here

Note: The .env file is gitignored and never committed. Each user provides their own API key.

Edit mcp_server/config.json to configure active airlines:

{
    "active_airlines": ["ryanair", "wizzair"],
    "available_airlines": ["ryanair", "wizzair", "google_flights"]
}

MCP Tools

Tool	Description
search_flights	Search flights between airports (origin, destination, date, return_date, adults)
list_active_airlines	Show currently active airlines
add_airline	Add an airline to active list
remove_airline	Remove an airline from active list
get_destinations	Get all available routes from an airport

Usage Example

Web UI

1. Enter IATA codes (BTS, STN, BCN...)
2. Select dates and passengers
3. Click "Search flights"
4. Optionally: check "Send email" to email results with booking links

AI Agent

You: Find the cheapest flight from Bratislava to London on July 15, 2026
Agent: [calls search_flights] Found RyanAir BTS->STN for 19.99 EUR...

Architecture

Browser → Flask (app.py) → MCP Server (server.py) → RyanAir/WizzAir API
                                                  → Google Flights
                        → Claude AI (optional, via SAP Hyperspace)

The web form searches without AI - directly calling MCP tools. The AI chat requires Claude (via SAP AI Proxy).

Testing

# Quick test (no AI needed):
python test_search.py

# Full agent test (8 use cases, requires AI Proxy):
python test_agent_prompts.py

Adding a New Airline

1. Create mcp_server/airlines/new_airline.py (inherit from BaseAirline)
2. Implement search_flights() and get_destinations()
3. Add import to mcp_server/airlines/__init__.py
4. Add to config.json → available_airlines
5. Add to server.py → get_airline_adapters()

Security

• API keys stored in .env (gitignored, never committed)
• Flask runs with debug=False
• CORS restricted to localhost
• Input validation and sanitization (IATA codes, dates, adults)
• Generic error messages (no internal info leaked)
• All logging to stderr (stdout reserved for MCP communication)

Tech Stack

• Python 3.12+
• MCP SDK 1.27+ (FastMCP)
• httpx (async HTTP)
• tenacity (retry with backoff)
• ryanair-py (RyanAir API)
• fast-flights (Google Flights)
• Flask + flask-cors (Web server)
• SAP Hyperspace AI Proxy (Claude)

License

MIT License - see LICENSE

Author

Robert Fodor - 2026