Paper Search MCP

Instructions

Implementation Reference

• paper_search_mcp/server.py:730-749 (handler)

  The `download_scihub` tool implementation in `server.py`, which is registered as an MCP tool using `@mcp.tool()`. It uses `SciHubFetcher` to download a PDF based on the provided identifier.

  @mcp.tool()
  async def download_scihub(
      identifier: str,
      save_path: str = "./downloads",
      base_url: str = "https://sci-hub.se",
  ) -> str:
      """Download paper PDF via Sci-Hub (optional fallback connector).
  
      Args:
          identifier: DOI, title, PMID, or paper URL.
          save_path: Directory to save the PDF.
          base_url: Sci-Hub mirror URL.
      Returns:
          Downloaded PDF path on success; error message on failure.
      """
      fetcher = SciHubFetcher(base_url=base_url, output_dir=save_path)
      result = await asyncio.to_thread(fetcher.download_pdf, identifier)
      if result:
          return result
      return "Sci-Hub download failed. Try DOI first, then title, or change mirror URL."