Agentic Travel Recommendations service for arrivia's white-label booking platform. Exposes both a REST API and an MCP server so AI Concierge agents can fetch member profiles and personalized recommendations, subject to partner-specific rules.
Two tools with clear, well-structured descriptions and proper input schemas. Both tools have verb-noun naming (get_*) and descriptions in the 150-200 char range, meeting baseline expectations. Schemas use Zod with type constraints (string.min(1), number.int().positive()). However, output schemas are not documented, responses are JSON-stringified without explicit field definitions. Error handling is present but generic (toolErrorResponse catches known errors but returns plain text). No tool annotations (readOnlyHint, idempotentHint) despite both being read-only operations. Parameter descriptions are adequate but lack format/constraint details (e.g., memberId format 'M-XXXX' is mentioned in description but not enforced in schema).
Retrieve a member's loyalty tier, partner affiliation, and last 5 bookings. Use this before calling get_recommendations if the agent needs to reason about the member (e.g. to write a personalized greeting).
Return personalized travel recommendations for a member. Automatically respects the partner's configuration: excluded categories are filtered out and the partner-imposed cap is honored. The response includes an `appliedRules` block so the agent can explain to the user why certain options may not appear.
Output schemas not documented. Responses are JSON-stringified text blocks without explicit field definitions. LLMs cannot plan downstream operations or extract structured data reliably.
No tool annotations despite both tools being read-only. Missing readOnlyHint would help agents understand these are safe to retry without side effects.
memberId parameter format is documented in description ('e.g. M-1001') but not enforced in schema. Should use pattern constraint or minLength/maxLength to prevent invalid IDs.
Error responses return plain text without structured guidance. 'Member not found' tells the agent nothing about recovery steps (e.g., 'Try search_members() first').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | <=2025-11-25 | v2 |
maxResults parameter in get_recommendations lacks explicit bounds. Should document minimum (1) and maximum (e.g., 100) to prevent LLM from passing absurd values.