Paper Search MCP Server

Instructions

INSTEAD (try in order):
1. read_scihub_paper(doi) - if published before 2023
2. read_semantic_paper(id) - last resort

Args:
    paper_id: PMID (unused).
    save_path: Unused.

Returns:
    Error message with alternatives.

Implementation Reference

• paper_find_mcp/server.py:283-298 (handler)

  The primary handler function decorated with @mcp.tool(), which registers it as an MCP tool and executes the logic by calling the generic _read helper for the 'pubmed' platform.

  @mcp.tool()
  async def read_pubmed_paper(paper_id: str, save_path: Optional[str] = None) -> str:
      """PubMed does NOT support direct paper reading.
      
      INSTEAD (try in order):
      1. read_scihub_paper(doi) - if published before 2023
      2. read_semantic_paper(id) - last resort
      
      Args:
          paper_id: PMID (unused).
          save_path: Unused.
      
      Returns:
          Error message with alternatives.
      """
      return await _read('pubmed', paper_id, save_path)

• paper_find_mcp/server.py:137-157 (helper)

  Generic helper function _read that all read_*_paper tools call. It retrieves the platform searcher from SEARCHERS dict and calls its read_paper method, handling NotImplementedError by returning the error message.

  async def _read(
      searcher_name: str, 
      paper_id: str, 
      save_path: Optional[str] = None
  ) -> str:
      """通用阅读函数"""
      if save_path is None:
          save_path = get_download_path()
      
      searcher = SEARCHERS.get(searcher_name)
      if not searcher:
          return f"Error: Unknown searcher {searcher_name}"
      
      try:
          return searcher.read_paper(paper_id, save_path)
      except NotImplementedError as e:
          return str(e)
      except Exception as e:
          logger.error(f"Read failed for {searcher_name}: {e}")
          return f"Error reading paper: {str(e)}"

• paper_find_mcp/academic_platforms/pubmed.py:338-350 (helper)

  The platform-specific read_paper implementation in PubMedSearcher class. Since PubMed does not host full-text papers, it returns an instructional message directing users to alternatives like DOI or PMC.

  def read_paper(self, paper_id: str, save_path: str) -> str:
      """尝试读取 PubMed 论文（不支持）
      
      Returns:
          str: 说明信息
      """
      return (
          "PubMed papers cannot be read directly through this tool. "
          "Only metadata and abstracts are available through PubMed's API. "
          "Please use the paper's DOI to access the full text, "
          "or check if the paper is available on PubMed Central (PMC)."
      )

• paper_find_mcp/server.py:283-298 (registration)

  The @mcp.tool() decorator registers this function as an MCP tool named 'read_pubmed_paper', with input schema inferred from the function signature.

  @mcp.tool()
  async def read_pubmed_paper(paper_id: str, save_path: Optional[str] = None) -> str:
      """PubMed does NOT support direct paper reading.
      
      INSTEAD (try in order):
      1. read_scihub_paper(doi) - if published before 2023
      2. read_semantic_paper(id) - last resort
      
      Args:
          paper_id: PMID (unused).
          save_path: Unused.
      
      Returns:
          Error message with alternatives.
      """
      return await _read('pubmed', paper_id, save_path)