Kortex Search demonstrates solid tool design with clear naming (verb_noun pattern), comprehensive parameter schemas, and detailed descriptions. All 8 tools follow consistent patterns with well-documented inputs. However, output schemas are not explicitly documented in the source code, and error handling guidance is minimal. Parameter descriptions are strong (averaging 80+ chars) and include constraints (enums, ranges). Tool composition is clean, each tool has a single responsibility. The server uses FastMCP with HTTP transport, enabling remote access. Main gaps: no visible output schema documentation, missing error recovery guidance, and no tool annotations (readOnlyHint/destructiveHint).

• Add explicit output schema documentation to each tool. Example: 'Returns: {results: [{title: string, url: string, snippet: string, source: string, score: float}], total_count: int, query: string}'.
• Add tool annotations via FastMCP's @mcp.tool(readOnlyHint=True) decorator to all 8 tools, signaling they are safe to call without side effects.
• Enhance error messages with actionable recovery steps. E.g., 'Rate limit on Semantic Scholar (429). Retry in 60s or use get_citations() with OpenAlex ID instead.'
• Document the 'fast set' default sources and explain CN tier opt-in in the search() description. Clarify which sources require API keys and which are free.
• Add pagination details to search_* descriptions: 'Results are not paginated; use limit (1-30) to control batch size. For large result sets, refine the query or use category filters.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry