Loading ToolBench
Auto-maps quiz input tags against catalog products to recommend targeted solutions.
The GenPark Custom Product Recommender is a minimal Python SDK with a single tool (match_quiz) that lacks fundamental MCP integration. The server is not an MCP server at all, it's a standalone Python class with no MCP protocol implementation, transport layer, or remote accessibility. The tool definition exists only in documentation form, not in a proper MCP schema. Input schema is present but parameter descriptions are minimal. Output schema is implicit (not formally documented). The codebase contains no evidence of MCP server registration, tool annotations, error handling patterns, or security considerations. This is a business logic library, not a production-grade agentic tool.
Auto-maps quiz input tags against catalog products to recommend targeted solutions.
Not an MCP server, standalone Python class with no MCP protocol transport or registration
Tool definition lacks MCP schema format, parameters are documented as raw JSON schema, not as JSON Schema objects with proper type definitions and descriptions
Parameter descriptions are absent or trivial. 'quiz_responses' lacks any hint of format or constraints. 'catalog' is described only as 'List of product catalog entries' with no documentation of required fields, expected structure, or validation rules
Output schema is not formally documented. Callers must infer from code that match_quiz returns {recommended_product_id, match_percentage}. No description of output fields, their types, or semantics for downstream tool chaining
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 26 | 2026-07-28+ | v2 |
No error handling or recovery guidance. The function silently returns None for recommended_product_id if catalog is empty. No validation of input structure, no error responses, no recovery hints for the LLM
Tool name 'match_quiz' does not start with a clear action verb. 'recommend_product_from_quiz' or 'search_catalog_by_quiz' would better convey intent and help LLMs select the right tool
No input validation. The function assumes quiz_responses values are stringifiable and catalog items have 'id' and 'tags' keys. Invalid input silently produces incorrect results rather than explicit errors
Confidence calculation is overly simplistic and misleading. Dividing max_matches by the number of questions (not unique answer tags) produces a percentage that does not reflect semantic relevance. No documentation of how confidence is computed