ai_classify

Instructions

Implementation Reference

• src/onion_mcp_server/tools/ai.py:301-338 (handler)

  Handler function for ai_classify tool. Builds a prompt based on the task (sentiment/topic/intent/custom) and calls llm_call to classify the input text. For custom tasks, it requires labels and returns an error if missing.

  elif name == "ai_classify":
      task    = a.get("task", "sentiment")
      explain = bool(a.get("explain", True))
      explain_str = "并简要说明理由" if explain else "，只输出分类结果，不需要解释"
  
      if task == "sentiment":
          prompt = (
              f"请对以下文本进行情感分析，判断情感倾向（正面/负面/中性）{explain_str}。\n\n"
              f"{a['text']}"
          )
      elif task == "topic":
          prompt = (
              f"请判断以下文本的主题类别（如：科技、政治、经济、体育、娱乐、教育、健康等）{explain_str}。\n\n"
              f"{a['text']}"
          )
      elif task == "intent":
          prompt = (
              f"请识别以下文本的用户意图（如：查询、投诉、购买、咨询、反馈等）{explain_str}。\n\n"
              f"{a['text']}"
          )
      elif task == "custom":
          labels = a.get("labels", [])
          if not labels:
              return [types.TextContent(type="text",
                  text="❌ task=custom 时必须提供 labels 参数")]
          labels_str = "、".join(f'"{line}"' for line in labels)
          prompt = (
              f"请将以下文本分类到这些类别之一：{labels_str}。\n"
              f"输出格式：分类结果{explain_str}。\n\n"
              f"{a['text']}"
          )
      else:
          raise ValueError(f"未知分类任务: {task}")
  
      reply = await llm_call(prompt)
      return [types.TextContent(type="text", text=reply)]
  
  raise ValueError(f"未知 ai 工具: {name}")

• src/onion_mcp_server/tools/ai.py:172-202 (schema)

  Schema definition for ai_classify tool. Defines input parameters: text (required), task (sentiment/topic/intent/custom), labels (for custom task), and explain (boolean).

  types.Tool(
      name="ai_classify",
      description="对文本进行分类，支持情感分析、主题分类、意图识别，或自定义分类标签。",
      inputSchema={
          "type": "object",
          "properties": {
              "text": {
                  "type":        "string",
                  "description": "要分类的文本",
              },
              "task": {
                  "type":        "string",
                  "description": "分类任务: sentiment（情感）/ topic（主题）/ intent（意图）/ custom（自定义）",
                  "enum":        ["sentiment", "topic", "intent", "custom"],
                  "default":     "sentiment",
              },
              "labels": {
                  "type":        "array",
                  "items":       {"type": "string"},
                  "description": "自定义分类标签（task=custom 时必填），如 [\"投诉\",\"咨询\",\"建议\"]",
                  "default":     [],
              },
              "explain": {
                  "type":        "boolean",
                  "description": "是否输出分类理由（默认 true）",
                  "default":     True,
              },
          },
          "required": ["text"],
      },
  ),

• src/onion_mcp_server/server.py:50-61 (registration)

  Registration: ai_classify is registered in the _HANDLERS routing table via iterating over AI_TOOLS and mapping each tool name to the handle_ai handler function.

  for _t in AI_TOOLS:     
      _HANDLERS[_t.name] = handle_ai
  for _t in CODE_TOOLS:   
      _HANDLERS[_t.name] = handle_code
  for _t in TEXT_TOOLS:   
      _HANDLERS[_t.name] = handle_text
  for _t in DATA_TOOLS:   
      _HANDLERS[_t.name] = handle_data
  for _t in WEB_TOOLS:    
      _HANDLERS[_t.name] = handle_web
  for _t in SYSTEM_TOOLS: 
      _HANDLERS[_t.name] = handle_system

• src/onion_mcp_server/tools/_llm.py:78-88 (helper)

  Helper llm_call function used by ai_classify handler to send the classification prompt to the LLM and return the response.

  async def llm_call(
      prompt: str,
      system: Optional[str] = None,
      temperature: float = 0.7,
  ) -> str:
      """单轮调用"""
      messages = []
      if system:
          messages.append({"role": "system", "content": system})
      messages.append({"role": "user", "content": prompt})
      return await llm_chat(messages, temperature=temperature)