MCP server for checking room availability and listing room types at a hostel using the Thais Hotel API
Two tools are directly registered with input schemas and descriptions in src/index.js. However, significant quality gaps undermine usability: (1) naming lacks clear verb-noun patterns, 'thais_list_room_types' uses a domain prefix unnecessarily when 'list_room_types' would suffice; (2) descriptions are adequate but lack WHEN-to-use context; (3) input schemas are present but incomplete, get_room_availability has good regex validation, but thais_list_room_types accepts empty input ({}) with no parameters, making discovery-based callers uncertain of expected output; (4) output schemas are not documented, responses are free-text wrapped in content arrays, not structured objects; (5) error handling is present but generic ('Error connecting to Thais API') without recovery guidance or error classification; (6) composition is weak, the lookup tool (thais_list_room_types) should perhaps be called automatically or its results cached if room IDs appear frequently in availability responses. Tool definitions are directly visible, so no inference penalty applies.
Check if rooms are available at the hostel for specific dates
Lookup tool to convert room_type_ids into human-readable names (e.g., discovering that ID 5 is a "Deluxe Suite"). Call this if you see IDs but don't know the room names.
thais_list_room_types accepts no input parameters ({}) but provides no documentation of expected output structure. Callers cannot discover the fields or shape of the response.
Output schemas are not documented. Both tools return free-text 'text' content wrapped in [{type: 'text', text: '...'}]. LLMs cannot infer downstream field names or plan chaining; a structured output schema (e.g., {room_type_id: int, label: string} for the lookup tool) would enable better composition.
Error messages are generic and provide no recovery guidance. 'Error connecting to Thais API: <error.message>' does not tell the LLM to retry, check credentials, or try an alternative approach.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 53 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 40 | - | v1 |
Tool naming uses domain prefix 'thais_' (e.g., 'thais_list_room_types') when a verb-noun pattern ('list_room_types') is clearer. LLMs parse intent from names first; 'thais_' adds noise.
Descriptions lack WHEN-to-use context. 'Check if rooms are available at the hostel for specific dates' explains WHAT but not WHEN an LLM should call this vs. a different search tool, or what happens next (e.g., 'Room IDs are opaque; call thais_list_room_types to learn room names').
Parameter descriptions in get_room_availability use example dates ('2025-03-15') in text. LLMs may reuse these literal values. Replace with formal constraints (pattern already declared in schema).