A simple MCP server for flight searches using Duffel API
Single tool with adequate schema but weak descriptions and no error guidance. Tool name 'get_flights' follows verb_noun convention (+). Input schema is present with types and descriptions for all 6 parameters (+). However, tool description is generic (59 chars, below 10-char minimum for actionable context) and lacks WHEN/WHY guidance. Parameter descriptions are minimal (e.g., 'The departure airport code or city' lacks format constraints). No output schema documented, LLM cannot predict response structure. Error handling returns unstructured dict with 'error' field but no recovery guidance. Missing pagination, result limits, and idempotency hints. No tool annotations (readOnlyHint present via Risk field but not in schema).
MCP tool for searching available flights between two locations. Returns both results and debug logs.
Tool description is too generic (59 chars). Lacks WHEN to use, prerequisites, and expected output structure. LLM cannot determine when to call get_flights vs. a hypothetical list_flights or search_flights_advanced.
No output schema documented. Response is unstructured dict with 'flights' (array), 'debug_log' (array), and optional 'error' (string). LLM cannot plan downstream calls or extract flight IDs for booking.
Parameter descriptions lack format constraints. 'departure_date' says 'ISO format' but no example or regex. 'cabin_class' has no enum, LLM may hallucinate invalid values like 'premium_economy' or 'first_class_deluxe'.
Error handling returns raw exception strings ('Error searching flights: ...') with no recovery guidance. LLM cannot determine if error is retryable, user-fixable, or fatal.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
No pagination or result limits. If Duffel API returns 500+ flight offers, response bloats context window. Tool description should state 'Returns up to 50 results' and accept limit/offset parameters.