A flight booking and management MCP server that provides tools for searching flights, managing bookings, and retrieving booking information
AmTrav MCP Server provides 5 flight management tools with explicit JSON Schema definitions visible in protocol.py. All tools have descriptions and properly typed input schemas with parameter descriptions. However, several critical gaps emerge: (1) output schemas are not documented, neither tool descriptions nor source code show what fields each tool returns, forcing LLMs to guess at response structure; (2) parameter descriptions lack actionable constraints (e.g., cabinClass says 'Economy, Business, or First (default: Economy)' but uses free-form strings instead of enums); (3) error handling is minimal, no guidance for recovery, no categorization of retryable vs fatal errors; (4) CreateBooking and ModifyBooking parameters reference flight_number and other domain concepts but no lookup tools exist to discover valid values (no search or list flight details tool); (5) no documentation of which tools modify state vs read-only beyond the Risk field; (6) tool composition issues, ModifyBooking and CancelBooking both require email + bookingId, but there's no guidance on how to obtain a bookingId from ListBookings. Naming is clear and verb-forward (SearchFlights, CreateBooking, etc.), and descriptions average ~80-120 chars, meeting baselines. Schemas are present and typed (object with properties, required arrays), but lack depth on constraints and output structure.
Cancel an existing flight booking
Create a new flight booking for a customer
Retrieve all flight bookings associated with a specific email address
Modify an existing booking's passenger name or flight number
Search for available flights based on origin, destination, date, and cabin class
Output schemas are completely undocumented. Tool descriptions do not specify what fields the LLM should expect in responses (e.g., SearchFlights returns flight list, but no field names, types, or structure are documented). LLMs cannot plan downstream calls or extract required data without seeing response shape.
Parameter 'cabinClass' in SearchFlights uses free-form string with example values in description ('Economy, Business, or First') instead of declaring as enum. LLMs may hallucinate invalid values like 'premium', 'first-class', or 'economy-plus'.
CreateBooking requires 'flightNumber' and 'totalPrice' as inputs, but no tool exists to discover valid flight numbers or calculate prices. SearchFlights returns results, but the LLM cannot determine which return fields map to CreateBooking's input parameters without response documentation. This breaks tool chaining.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 56 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 47 | 2024-11-05+ | v1 |
ModifyBooking and CancelBooking both require 'bookingId', but ListBookings description does not specify that it returns bookingId in its response. Without documented response structure, the LLM cannot extract the booking ID to pass to downstream tools.
No error handling documentation. Tool descriptions do not explain failure modes, which errors are retryable, or what the LLM should do if a booking is not found or a flight is overbooked. Error responses from the service are not shown.
Risk field (READ_ONLY, WRITE, REVERSIBLE, DESTRUCTIVE) is metadata but not part of tool descriptions. LLMs cannot infer from descriptions alone which tools are safe to retry vs which have irreversible consequences. Risk should be conveyed in tool descriptions or via tool annotations.
Parameter 'departureDate' in SearchFlights says 'YYYY-MM-DD format' in description but does not validate or constrain dates (e.g., no min/max dates, no validation that date is in the future). LLMs may pass invalid dates like '1999-01-01' or '2099-12-31' without guidance.
Parameter 'origin' and 'destination' in SearchFlights accept free-form strings but examples suggest IATA airport codes ('JFK', 'LAX'). No enum of valid codes, no format constraint, no guidance on 3-letter vs full name. LLMs may pass invalid or ambiguous values.
ListBookings description says 'Retrieve all flight bookings associated with a specific email address' but does not document pagination (no limit, offset, or page_size parameters). If a user has 1000+ bookings, returning all may exhaust context or timeout. Should include limit/offset and document max results.
ModifyBooking has optional parameters 'newPassengerName' and 'newFlightNumber' but description does not state that at least one is required. If LLM passes neither (only email + bookingId), the call may silently fail or update nothing. Document mutual constraint or enforce at input validation layer.