Scholar MCP Server

Instructions

Implementation Reference

• scholar_mcp_server.py:85-204 (handler)

  The `paper_ai_analyze` tool handler implementation. It uses Crossref to fetch paper metadata, optionally downloads and extracts text from the PDF, and uses an AI API to analyze the content.

  def paper_ai_analyze(doi: str) -> str:
      """使用 AI 分析论文，返回核心贡献、研究方法、关键发现等。
  
      支持任意 OpenAI 兼容 API（通过 AI_API_BASE / AI_API_KEY / AI_MODEL 环境变量配置）。
      如果能下载到 PDF，会提取全文进行深度分析；否则退回到 abstract 分析。
  
      Args:
          doi: 论文的 DOI，例如 "10.1109/tim.2021.3106677"
  
      Returns:
          AI 分析结果 JSON
      """
      if not AI_API_KEY:
          return json.dumps({"success": False, "error": "AI_API_KEY (or DS_KEY) not set, AI analysis unavailable"})
  
      try:
          # 获取论文元数据
          r = requests.get(f"{CROSSREF}/{requests.utils.quote(doi)}", timeout=10)
          if r.status_code != 200:
              return json.dumps({"success": False, "error": f"DOI not found: {doi}"})
  
          item = r.json().get("message", {})
          title = (item.get("title") or [""])[0]
          authors = ", ".join(
              f"{a.get('given', '')} {a.get('family', '')}"
              for a in (item.get("author") or [])
          )
          journal = (item.get("container-title") or [""])[0]
          year = ""
          if item.get("published") and item["published"].get("date-parts"):
              year = str(item["published"]["date-parts"][0][0])
          abstract = (
              (item.get("abstract") or "")
              .replace("<jats:p>", "").replace("</jats:p>", "")
              .strip()
          )
  
          # 尝试获取全文：下载 PDF → 提取文本
          full_text = ""
          analysis_mode = "abstract_only"
          import tempfile
          try:
              tmp_dir = tempfile.mkdtemp(prefix="scholar_mcp_")
              dl_result = download_paper(doi, tmp_dir)
              if dl_result.get("success") and dl_result.get("path"):
                  pdf_result = extract_pdf_text(dl_result["path"], max_pages=20, max_chars=12000)
                  if pdf_result.get("success"):
                      full_text = pdf_result["text"]
                      analysis_mode = f"full_text ({pdf_result['pages']} pages, {pdf_result['chars']} chars, {pdf_result['method']})"
          except Exception:
              pass
  
          # 构建 prompt
          if full_text:
              prompt = f"""You are an academic research assistant. Analyze this paper thoroughly in Chinese based on the full text provided.
  
  Title: {title}
  Authors: {authors}
  Journal: {journal} ({year})
  
  --- FULL TEXT ---
  {full_text}
  --- END ---
  
  Provide a thorough analysis:
  1. **核心贡献**: 一句话概括
  2. **研究背景**: 为什么做这个研究（2-3句）
  3. **研究方法**: 详细描述使用的方法和技术路线（3-5句）
  4. **关键发现**: 3-5个具体的定量或定性结果
  5. **创新点**: 与现有研究的具体区别
  6. **局限性**: 2-3点
  7. **应用前景**: 1-2句"""
              max_tokens = 1500
          else:
              prompt = f"""You are an academic research assistant. Analyze this paper concisely in Chinese.
  
  Title: {title}
  Authors: {authors}
  Journal: {journal} ({year})
  {"Abstract: " + abstract if abstract else ""}
  
  Provide:
  1. **核心贡献**: 一句话概括
  2. **研究方法**: 2-3句
  3. **关键发现**: 2-3个要点
  4. **创新点**: 与现有研究的区别
  5. **局限性**: 1-2点"""
              max_tokens = 800
  
          # 调用 AI API
          api_url = f"{AI_API_BASE.rstrip('/')}/chat/completions"
          ar = requests.post(
              api_url,
              headers={
                  "Authorization": f"Bearer {AI_API_KEY}",
                  "Content-Type": "application/json",
              },
              json={
                  "model": AI_MODEL,
                  "messages": [{"role": "user", "content": prompt}],
                  "temperature": 0.3,
                  "max_tokens": max_tokens,
              },
              timeout=120,
          )
          analysis = ar.json()["choices"][0]["message"]["content"]
  
          return json.dumps({
              "success": True,
              "doi": doi,
              "title": title,
              "authors": authors,
              "journal": f"{journal} ({year})",
              "model": AI_MODEL,
              "analysis_mode": analysis_mode,
              "analysis": analysis,
          }, ensure_ascii=False, indent=2)
  
      except Exception as e:
          return json.dumps({"success": False, "error": str(e)})

• scholar_mcp_server.py:84-84 (registration)

  Registration of the `paper_ai_analyze` tool using the @mcp.tool decorator.

  @mcp.tool()