ask_rag
Query a knowledge base to get answers from processed documents using semantic search and vector retrieval technology.
Instructions
向 RAG 知识库提问，并根据存储的信息返回答案。 使用场景：
询问特定主题或概念
请求解释或定义
从处理过的文档中获取信息
基于学习的文本或文档获取答案
参数： query: 要向知识库提出的问题或查询。
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes |
Implementation Reference
- src/mcp_rag/mcp_server.py:27-71 (registration)Registers the MCP tool named 'rag_ask' (likely the intended 'ask_rag') with its input schema definition including parameters for query, mode, collection, limit, and threshold.
@self.server.list_tools() async def list_tools() -> List[Tool]: """列出可用的MCP工具。""" return [ Tool( name="rag_ask", description="向RAG知识库提问查询信息", inputSchema={ "type": "object", "properties": { "query": { "type": "string", "description": "搜索查询" }, "mode": { "type": "string", "enum": ["raw", "summary"], "description": "检索模式", "default": "raw" }, "collection": { "type": "string", "description": "要搜索的集合名称", "default": "default" }, "limit": { "type": "integer", "description": "最大结果数量", "default": 5, "minimum": 1, "maximum": 20 }, "threshold": { "type": "number", "description": "相似度阈值", "default": 0.7, "minimum": 0.0, "maximum": 1.0 } }, "required": ["query"] } ) ] - src/mcp_rag/mcp_server.py:31-69 (schema)Input schema for the 'rag_ask' tool defining parameters: query (required string), mode (raw/summary), collection (default 'default'), limit (1-20, default 5), threshold (0.0-1.0, default 0.7).
Tool( name="rag_ask", description="向RAG知识库提问查询信息", inputSchema={ "type": "object", "properties": { "query": { "type": "string", "description": "搜索查询" }, "mode": { "type": "string", "enum": ["raw", "summary"], "description": "检索模式", "default": "raw" }, "collection": { "type": "string", "description": "要搜索的集合名称", "default": "default" }, "limit": { "type": "integer", "description": "最大结果数量", "default": 5, "minimum": 1, "maximum": 20 }, "threshold": { "type": "number", "description": "相似度阈值", "default": 0.7, "minimum": 0.0, "maximum": 1.0 } }, "required": ["query"] } ) - src/mcp_rag/mcp_server.py:72-137 (handler)The handler function for MCP tool calls. For 'rag_ask', it extracts parameters, embeds the query using embedding model, searches the vector database, formats and returns search results as text content. Supports 'raw' and 'summary' modes (summary not fully implemented).
@self.server.call_tool() async def call_tool(name: str, arguments: Dict[str, Any]) -> Sequence[types.TextContent]: """调用MCP工具。""" if name == "rag_ask": try: logger.info(f"开始处理RAG检索请求: {arguments.get('query', 'unknown')}") # 获取组件 vector_db = await get_vector_database() embedding_model = await get_embedding_model() # 提取参数 query = arguments["query"] mode = arguments.get("mode", "raw") collection = arguments.get("collection", "default") limit = arguments.get("limit", 5) threshold = arguments.get("threshold", 0.7) logger.info(f"编码查询: {query}") # 编码查询 query_embedding = await embedding_model.encode_single(query) logger.info(f"搜索数据库，集合: {collection}, 限制: {limit}") # 搜索数据库 search_results = await vector_db.search( query_embedding=query_embedding, collection_name=collection, limit=limit, threshold=threshold ) # 格式化结果 if not search_results: logger.info("未找到相关文档") return [types.TextContent( type="text", text=f"为查询 '{query}' 未找到相关文档" )] logger.info(f"找到 {len(search_results)} 个相关文档") result_text = f"为查询 '{query}' 找到 {len(search_results)} 个相关文档\n\n" for i, result in enumerate(search_results, 1): result_text += f"[{i}] 相似度: {result.score:.3f}\n" result_text += f"内容: {result.document.content}\n" if result.document.metadata.get("source"): result_text += f"来源: {result.document.metadata['source']}\n" result_text += "\n" # 对于摘要模式，添加摘要生成 if mode == "summary": result_text += "\n--- 摘要模式 ---\n" result_text += "摘要生成功能尚未实现。\n" logger.info("RAG检索完成") return [types.TextContent(type="text", text=result_text)] except Exception as e: logger.error(f"工具调用失败: {e}") return [types.TextContent( type="text", text=f"检索过程中出错: {str(e)}" )] else: raise ValueError(f"未知工具: {name}")