滴个龟龟 — 龟类识别+饲养+品相+血统 API。MCP (Model Context Protocol) JSON-RPC 2.0 over HTTP server providing turtle species identification, care parameters, genetics calculation, provenance verification, and appraisal tools.
Server has 6 tools with basic schemas and Chinese descriptions. Critical gaps: (1) All descriptions are in Chinese only, making them inaccessible to English-speaking LLMs and violating pattern:tool-description baseline of 194 chars avg in English; (2) Parameter descriptions are minimal or missing context (e.g., 'genes' accepts both string and array but no guidance on format); (3) No output schemas documented, LLMs cannot predict response structure for chaining; (4) Error handling is minimal, most tools return {'error': '...'} with no recovery guidance; (5) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being READ_ONLY; (6) Tool names are verb-noun but lack clarity on when to use each (e.g., identify_turtle vs search_species both find species). Baseline: 549 production tools average 18-char names and 194-char descriptions in English. This server's descriptions average ~40 chars in Chinese, insufficient for LLM reasoning.
价值预估
基因计算器 — 调用 genecalc.py 核心逻辑
获取品种完整档案
拍照识龟 — 调本地推理服务
搜索品种 — 支持学名/中文名/俗称/分组
扫码验证爬宠身份
All tool and parameter descriptions are in Chinese only. LLMs trained primarily on English cannot reason about tool selection or parameter constraints. Descriptions must be in English (or bilingual with English primary) to be usable by most LLM agents.
No output schemas documented. Tools return complex nested objects (e.g., get_profile returns species with care_params, price_cny, aliases) but LLMs have no schema to predict structure. This breaks tool chaining and forces agents to guess field names.
Parameter 'genes' in estimate_value accepts both string and array but provides no format guidance. Description says '支持逗号分隔字符串或列表' (supports comma-separated string or list) but LLMs cannot parse Chinese constraints. Specify format in English with examples: 'Comma-separated string (e.g., "het-red,het-orange") or JSON array'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Error handling returns bare error objects with no recovery guidance. E.g., identify_turtle returns {'error': '推理服务不可用: ...'} with no actionable next step. Per pattern:recovery-guide, errors must tell LLM what to do: 'Inference service unavailable. Try again in 30s or use search_species() to manually find the species.'
No tool annotations despite all tools being READ_ONLY. Missing readOnlyHint annotation prevents agents from understanding which tools are safe to call speculatively. Add 'readOnlyHint: true' to all 6 tools.