Scholar MCP Server

Instructions

Implementation Reference

• downloader.py:244-270 (handler)

  The actual logic for paper_download is implemented here in download_paper, which coordinates multiple download sources.

  def download_paper(doi: str, output_dir: str = ".") -> dict:
      """下载论文的主入口，按优先级尝试多个来源
  
      优先级：Unpaywall (合法OA) → arXiv → Sci-Hub (自定义) → Sci-Hub (scidownl)
      """
      os.makedirs(output_dir, exist_ok=True)
      output = os.path.join(output_dir, _safe_name(doi))
  
      # 已存在则跳过
      if os.path.exists(output):
          size_mb = os.path.getsize(output) / 1024 / 1024
          return {
              "success": True,
              "doi": doi,
              "path": os.path.abspath(output),
              "size_mb": round(size_mb, 2),
              "source": "cached",
          }
  
      # 1. Unpaywall (合法 OA)
      oa_url = _try_unpaywall(doi)
      if oa_url:
          try:
              r = requests.get(oa_url, timeout=30, allow_redirects=True)
              if r.content[:5] == b"%PDF-":
                  with open(output, "wb") as f:
                      f.write(r.content)

• scholar_mcp_server.py:37-49 (registration)

  The paper_download tool is registered as an MCP tool here.

  @mcp.tool()
  def paper_download(doi: str, output_dir: str = ".") -> str:
      """通过 DOI 下载一篇论文 PDF（本地多源下载：Unpaywall → arXiv → Sci-Hub）。
  
      Args:
          doi: 论文的 DOI，例如 "10.1109/tim.2021.3106677"
          output_dir: 保存 PDF 的目录路径，默认为当前目录
  
      Returns:
          下载结果的 JSON 字符串，包含 success, doi, path, size_mb, source 等字段
      """
      result = download_paper(doi, output_dir)
      return json.dumps(result, ensure_ascii=False, indent=2)