Smart Search — Yuque Team Knowledge Base Search & Q&A

Search across group Yuque knowledge bases using natural language, read relevant documents, and synthesize a clear answer with references. Scoped to team/group repositories.

When to Use

• User asks a question that might be answered by their team's Yuque docs

• User wants to find specific information in group knowledge bases

• User says "搜一下团队文档", "search group Yuque", "团队知识库里有没有..."

Required MCP Tools

All tools are from the yuque-mcp server:

• yuque_list_repos — List group knowledge bases to determine search scope

• yuque_search — Search documents by keyword

• yuque_get_doc — Read full document content by slug/id

Workflow

Step 1: Identify Team Scope

First, get the list of group knowledge bases to understand the search scope:

Tool: yuque_list_repos Parameters: login: "<group_login>" type: "group"

If the user hasn't specified a group login, ask: "请告诉我团队的语雀团队标识（group login），我来搜索团队知识库。"

Step 2: Extract Search Keywords

From the user's natural language query, extract 1-3 concise Chinese keywords. Yuque search works best with short, specific terms.

Examples:

• "团队的部署流程是什么？" → keywords: 部署流程

• "我们的 API 规范文档" → keywords: API 规范

• "上次技术评审的结论" → keywords: 技术评审

Step 3: Search Documents

Call yuque_search with the extracted keywords:

Tool: yuque_search Parameters: query: "<keywords>" type: "doc"

If the search returns no results:

• Try alternative keywords (synonyms, broader terms)

• Try splitting compound terms (e.g., "部署流程" → "部署" or "流程")

• If still no results, tell the user honestly: "在团队语雀知识库中未找到相关文档，建议尝试其他关键词或确认文档是否存在。"

Step 4: Filter and Rank Results

From the search results, select the top 3-5 most relevant documents based on:

• Title relevance to the query

• Document update time (prefer recent)

• Repository context (prioritize results from group repos identified in Step 1)

Filter out results that don't belong to the team's knowledge bases.

Step 5: Read Document Content

For each selected document, fetch the full content:

Tool: yuque_get_doc Parameters: repo_id: "<namespace>" # e.g., "my-team/tech-docs" (from search result's book.namespace) doc_id: "<slug>" # e.g., "api-spec" (from search result's target.slug)

Note: The search result contains nested data. Extract these key fields:

• target.slug → use as doc_id

• target.book.namespace → use as repo_id

• target.title → document title

• target.updated_at → last update time

Read up to 3 documents. If the first document fully answers the question, you may skip the rest.

Step 6: Synthesize and Respond

Compose the answer in the following format:

回答

[直接回答用户的问题，2-4 句话，简洁明了]

关键要点

• 要点 1：[从文档中提取的关键信息]
• 要点 2：[从文档中提取的关键信息]
• 要点 3：[从文档中提取的关键信息]

参考文档

1. 文档标题 — 来自「知识库名称」，更新于 YYYY-MM-DD
2. 文档标题 — 来自「知识库名称」，更新于 YYYY-MM-DD

Guidelines

• Always answer in the same language the user used (Chinese or English)

• Quote specific content from documents when relevant — use > blockquotes

• If documents contain conflicting information, note the discrepancy and mention which doc is more recent

• If the answer is only partially found, say what you found and what's missing

• Never fabricate information not present in the documents

• Include document links so the user can read the full source

• Only return results from group knowledge bases — filter out personal docs

• This skill searches group knowledge bases — for personal docs, use the smart-search skill in the yuque-personal plugin

Error Handling

Situation Action

yuque_list_repos fails Check if group login is correct and group Token is configured

yuque_search returns empty Try alternative keywords, then inform user

yuque_get_doc fails (404) Skip this doc, note it may have been deleted

yuque_get_doc fails (403) Tell user they may lack permission; check group Token scope

API timeout Retry once, then inform user of connectivity issue

Too many results Focus on top 3 by relevance, mention there are more results available

Team Token not configured Inform user that group search requires a team-level Token