中国威士忌知识库 MCP 服务器,提供酒厂查询、产品推荐、品鉴数据漏斗、内容生成等功能。采用诚实数据标注(已核实/待确认/待采集)和真人品鉴聚合机制。
Server defines 9 tools with Chinese descriptions and basic schemas. All tools have names starting with action verbs (list_, get_, search_, submit_, recommend_, generate_). Schemas are present and typed for most tools. Error handling is minimal. Tool composition is reasonable but some tools conflate multiple concerns (e.g., generate_content mixes content generation with data validation).
内容生成:从国威知识库取真实数据,按格式生成内容骨架(选题/长文/小红书/短视频口播),待核实项会标出
查某家国产威士忌酒厂的完整档案(产区/工艺/风土/产品/可信度)
查某家酒厂的风味图谱(主导风味词 + 官方/社区来源),含待采集诚实标注
列出国产威士忌酒厂,可按产区过滤(region 用产业区 code)
查某家酒厂的产品线与价格带
列出中国威士忌主要产区萌芽体系
选酒导购:按预算+口味+场景,从国威知识库推荐酒款(含可信度与诚实声明)
Descriptions too brief for LLM selection. list_regions (列出中国威士忌主要产区萌芽体系) is ~25 chars; rubric baseline is 194 chars avg. LLMs cannot infer when to call vs similar tools.
Output schemas not formally documented. Tools return structured objects (summarize(), detail(), etc.) but no JSON Schema or TypeScript interface is exposed to the MCP client. LLMs cannot plan downstream calls without knowing response fields.
Parameter descriptions lack format/constraint details. 'region' param in list_distilleries says '产业区 code,如 qionglai' but does not specify: valid values, case sensitivity, or how to discover valid codes. LLMs will guess.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
在国威知识库中检索(匹配酒厂名/产区/风格/主导风味词/故事)
提交一条品鉴笔记(数据漏斗)。仅采信真人来源;AI 不据此给任何权威结论。
Error handling minimal. submit_tasting_note returns {accepted: false, error: '...'} but does not guide recovery. E.g., if distilleryId not found, should suggest calling list_distilleries() first.
generate_content conflates content generation with data validation. Tool mixes 'take real data from knowledge base' with 'generate content skeleton', should split into separate tools (fetch_content_data + generate_content_skeleton) for composability.