Production-grade client for interfacing with GenPark Autopilot Search API. Handles dynamic payload construction, ranking logic, response sanitization, and automatic exponential backoff retries for robust server operation.
This MCP server exposes a single search tool with a reasonable schema and basic description, but falls significantly short of production-grade quality. The tool description (89 chars) is below the baseline (194 chars avg for A+ tools) and lacks critical context about when/why to use it vs alternatives. While the input schema is present with proper types, parameter descriptions are minimal and do not explain constraints, formats, or error conditions. The tool performs read-only search operations but lacks idempotency guarantees, pagination support for large result sets, and structured error guidance. The output schema is not formally documented in the tool definition, it is only visible in the example_usage.py file, meaning an LLM evaluating the tool cannot determine expected response structure without reverse-engineering from examples. No tool annotations (readOnlyHint, idempotentHint) are declared. Security review finds no obvious credential leaks, but the mock mode fallback without strict API key validation could mask production issues.
Queries the GenPark indexing service, resolves results, and generates structural citations.
Tool description is too short (89 chars, baseline 194). Lacks WHEN/WHY/WHAT context required for LLM tool selection. Does not explain how this search differs from other search tools or when to use it instead of alternatives.
Output schema is not declared in the tool definition. The response structure (ranked_results, citations, web_result_url) is only visible in example_usage.py, not in the tool metadata. LLMs cannot plan downstream operations without a documented schema.
Parameter 'query' description lacks constraint information. Does not specify minimum length, character restrictions, or format requirements. 'Cannot be empty' is mentioned but not formalized as a min-length constraint.
Parameter 'max_results' description missing actionable constraints. Does not specify the valid range (e.g., 1 - 100) or performance implications of large values. Baseline shows 100% of A+ tools document numeric bounds.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
No pagination support documented. If search returns 'ranked_results' as an unbounded list, large result sets could exhaust context windows. No mention of offset, page, or limit enforcement in tool description.
No error handling guidance in description. Does not explain what happens when the API is unavailable, the query is invalid, or no results are found. The code has a GenParkSearchError exception but this is not surfaced in tool docs.
Tool annotations missing. The 'search' tool is read-only and idempotent but does not declare readOnlyHint or idempotentHint. LLMs cannot determine retry safety or caching eligibility without these annotations.
Mock mode behavior not documented in tool description. The tool silently falls back to simulation if API_KEY is missing. LLMs and users have no way to know whether results are real or synthetic without reading the source code.