sjtu_cheap_task

Instructions

Implementation Reference

• src/sjtu_mcp/server.py:112-148 (handler)

  The async function `sjtu_cheap_task` is the handler for the 'sjtu_cheap_task' MCP tool. It routes common low-risk jobs (summarize, rewrite, classify, extract, etc.) by building a routing prompt and calling the LLM API. Supports both text-only and image+text inputs, with model selection based on task type (reasoning models for 'reason'/'analyze'/'plan', text model otherwise).

  async def sjtu_cheap_task(
      task: str,
      content: str,
      image_path: str | None = None,
      image_url: str | None = None,
      model: str | None = None,
  ) -> str:
      """Route common low-risk jobs like summarize, rewrite, classify, and extract."""
      task_name = task.strip().lower()
      routing_prompt = (
          "You are handling a low-risk utility task. Be concise, accurate, and structured.\n"
          f"Task: {task_name}\n"
          "Return the useful result directly."
      )
  
      if image_path or image_url:
          response = await client.chat(
              model=model or settings.default_vision_model,
              messages=_build_vision_messages(
                  f"{routing_prompt}\n\nInput:\n{content}",
                  image_path=image_path,
                  image_url=image_url,
                  system_prompt=None,
              ),
              temperature=0.1,
          )
          return _extract_text(response)
  
      selected_model = model or (
          settings.default_reasoning_model if task_name in {"reason", "analyze", "plan"} else settings.default_text_model
      )
      response = await client.chat(
          model=selected_model,
          messages=_build_text_messages(f"{routing_prompt}\n\nInput:\n{content}", None),
          temperature=0.1,
      )
      return _extract_text(response)

• src/sjtu_mcp/server.py:111-111 (registration)

  The `@mcp.tool()` decorator on line 111 registers `sjtu_cheap_task` as an MCP tool with the FastMCP server instance `mcp`.

  @mcp.tool()

• src/sjtu_mcp/server.py:15-29 (helper)

  Helper function `_extract_text` used by the handler to extract text from the API response dictionary.

  def _extract_text(response: dict[str, Any]) -> str:
      choices = response.get("choices", [])
      if not choices:
          return "No choices returned."
      message = choices[0].get("message", {})
      content = message.get("content", "")
      if isinstance(content, str):
          return content
      if isinstance(content, list):
          chunks: list[str] = []
          for item in content:
              if isinstance(item, dict) and item.get("type") == "text":
                  chunks.append(str(item.get("text", "")))
          return "\n".join(chunk for chunk in chunks if chunk)
      return str(content)

• src/sjtu_mcp/server.py:40-60 (helper)

  Helper function `_build_vision_messages` used when the handler processes image inputs.

  def _build_vision_messages(
      prompt: str,
      *,
      image_path: str | None,
      image_url: str | None,
      system_prompt: str | None,
  ) -> list[dict[str, Any]]:
      if not image_path and not image_url:
          raise ValueError("Provide either image_path or image_url.")
  
      image_source = image_url or image_path_to_data_url(image_path or "")
      user_content = [
          {"type": "text", "text": prompt},
          {"type": "image_url", "image_url": {"url": image_source}},
      ]
  
      messages: list[dict[str, Any]] = []
      if system_prompt:
          messages.append({"role": "system", "content": system_prompt})
      messages.append({"role": "user", "content": user_content})
      return messages

• src/sjtu_mcp/server.py:32-37 (helper)

  Helper function `_build_text_messages` used by the handler for text-only inputs.

  def _build_text_messages(prompt: str, system_prompt: str | None) -> list[dict[str, Any]]:
      messages: list[dict[str, Any]] = []
      if system_prompt:
          messages.append({"role": "system", "content": system_prompt})
      messages.append({"role": "user", "content": prompt})
      return messages