Paper Search MCP Server

Instructions

Implementation Reference

• paper_find_mcp/server.py:731-816 (handler)

  Primary handler and registration (@mcp.tool()) for the 'search_repec' MCP tool. Defines input schema via type hints and extensive docstring. Delegates execution to _search('repec', ...) which invokes RePECSearcher.search().
  @mcp.tool() async def search_repec( query: str, max_results: int = 10, year_from: Optional[int] = None, year_to: Optional[int] = None, search_field: str = 'all', sort_by: str = 'relevance', doc_type: str = 'all', series: Optional[str] = None, ) -> List[Dict]: """Search economics papers on RePEc/IDEAS - the largest open economics bibliography. USE THIS TOOL WHEN: - Searching for ECONOMICS research (macro, micro, finance, econometrics) - You need working papers from NBER, Federal Reserve, World Bank, etc. - You want to find papers by JEL classification - Searching for economic policy analysis COVERAGE: 4.5M+ items including: - Working Papers: NBER, Fed banks, ECB, IMF, World Bank - Journal Articles: AER, JPE, QJE, Econometrica, etc. - Books and Book Chapters SEARCH SYNTAX: - Boolean: + for AND, | for OR, ~ for NOT (e.g., 'money ~liquidity') - Phrase: use double quotes (e.g., '"monetary policy"') - Author(Year): e.g., 'Acemoglu (2019)' or 'Kydland Prescott (1977)' - Synonyms: automatic (labor=labour, USA=United States) - Word stemming: automatic (find matches finds, finding, findings) LIMITATION: RePEc provides metadata only, not full PDFs. PDFs are hosted at original institutions (often freely available). Args: query: Search terms with optional boolean operators. max_results: Number of results (default: 10). year_from: Optional start year filter (e.g., 2020). year_to: Optional end year filter (e.g., 2025). search_field: Where to search, one of: - 'all': Whole record (default) - 'abstract': Abstract only - 'keywords': Keywords only - 'title': Title only - 'author': Author only sort_by: How to sort results, one of: - 'relevance': Most relevant (default) - 'newest': Most recent first - 'oldest': Oldest first - 'citations': Most cited first - 'recent_relevant': Recent and relevant - 'relevant_cited': Relevant and cited doc_type: Document type filter, one of: - 'all': All types (default) - 'articles': Journal articles - 'papers': Working papers (NBER, Fed, etc.) - 'chapters': Book chapters - 'books': Books - 'software': Software components series: Institution/journal series to search within, one of: - Institutions: 'nber', 'imf', 'worldbank', 'ecb', 'bis', 'cepr', 'iza' - Federal Reserve: 'fed', 'fed_ny', 'fed_chicago', 'fed_stlouis' - Top 5 Journals: 'aer', 'jpe', 'qje', 'econometrica', 'restud' - Other journals: 'jfe', 'jme', 'aej_macro', 'aej_micro', 'aej_applied' Returns: List of paper dicts with: paper_id (RePEc handle), title, authors, abstract, published_date, url, categories (JEL codes). Example: search_repec('inflation', series='nber') # Search NBER only search_repec('causal inference', series='aer', sort_by='newest') search_repec('machine learning', series='fed', year_from=2020) """ kwargs = { 'year_from': year_from, 'year_to': year_to, 'search_field': search_field, 'sort_by': sort_by, 'doc_type': doc_type, 'series': series, } # Remove None values kwargs = {k: v for k, v in kwargs.items() if v is not None} return await _search('repec', query, max_results, **kwargs)
• paper_find_mcp/academic_platforms/repec.py:414-570 (helper)

  Core implementation of the RePEc search logic in RePECSearcher.search(). Constructs advanced search query for IDEAS/RePEc, scrapes results via POST request and BeautifulSoup parsing, extracts paper metadata including RePEc handles, titles, URLs, and years.
  def search( self, query: str, max_results: int = 10, year_from: Optional[int] = None, year_to: Optional[int] = None, search_field: str = 'all', sort_by: str = 'relevance', doc_type: str = 'all', series: Optional[str] = None, ) -> List[Paper]: """搜索 RePEc/IDEAS 论文 支持 IDEAS 高级搜索的所有选项。 搜索语法提示: - 布尔搜索: + 表示 AND, | 表示 OR, ~ 表示 NOT - 短语搜索: 使用双引号, 例如 "monetary policy" - 作者(年份): 例如 "Acemoglu (2019)" - 自动同义词: labor=labour, USA=United States - 词干提取: find 匹配 finds, finding, findings Args: query: 搜索关键词 (支持布尔运算符) max_results: 最大返回数量 (默认 10) year_from: 起始年份 (可选, 如 2020) year_to: 结束年份 (可选, 如 2025) search_field: 搜索字段, 可选值: - 'all': 全部字段 (默认) - 'abstract': 仅摘要 - 'keywords': 仅关键词 - 'title': 仅标题 - 'author': 仅作者 sort_by: 排序方式, 可选值: - 'relevance': 相关性 (默认) - 'newest': 最新发布 - 'oldest': 最早发布 - 'citations': 被引用最多 - 'recent_relevant': 最新且相关 - 'relevant_cited': 相关且被引用 doc_type: 文档类型, 可选值: - 'all': 所有类型 (默认) - 'articles': 期刊文章 - 'papers': 工作论文 (NBER, Fed 等) - 'chapters': 书籍章节 - 'books': 书籍 - 'software': 软件组件 series: 研究机构/期刊系列, 可选值: - 机构: 'nber', 'imf', 'worldbank', 'ecb', 'bis', 'cepr', 'iza' - 美联储: 'fed', 'fed_ny', 'fed_chicago', 'fed_stlouis' - 期刊: 'aer', 'jpe', 'qje', 'econometrica', 'restud' - 其他: 'jfe', 'jme', 'aej_macro', 'aej_micro', 'aej_applied' Returns: List[Paper]: 论文列表 Example: >>> papers = searcher.search("artificial intelligence", max_results=5) >>> papers = searcher.search('"monetary policy" +inflation', sort_by='newest') >>> papers = searcher.search("inflation", series='nber') # 仅搜索 NBER >>> papers = searcher.search("causal", series='aer') # 仅搜索 AER """ if not query or not query.strip(): return [] papers = [] seen_urls = set() # 避免重复 try: # 获取参数值 wf = self.SEARCH_FIELDS.get(search_field, self.SEARCH_FIELDS['all']) s = self.SORT_OPTIONS.get(sort_by, self.SORT_OPTIONS['relevance']) # 处理 ul 参数: series 优先于 doc_type if series: ul = self.SERIES.get(series, series) # 支持直接传入 handle else: ul = self.DOC_TYPES.get(doc_type, self.DOC_TYPES['all']) # 构建 POST 数据 data = { 'q': query, 'wf': wf, # 搜索字段 's': s, # 排序方式 'form': 'extended', 'wm': 'wrd', 'dt': 'range', } # 添加 ul 参数 (仅当非空时) if ul: data['ul'] = ul # 添加日期范围 if year_from: data['db'] = f'01/01/{year_from}' if year_to: data['de'] = f'12/31/{year_to}' # 随机延迟，避免被封 time.sleep(random.uniform(0.5, 1.5)) # 发送 POST 请求 response = self.session.post( self.SEARCH_URL, data=data, timeout=self.timeout ) if response.status_code != 200: logger.error(f"RePEc search failed with status {response.status_code}") return [] # 解析 HTML soup = BeautifulSoup(response.text, 'html.parser') # 查找所有论文链接 for link in soup.find_all('a', href=True): if len(papers) >= max_results: break href = link.get('href', '') # 检查是否为论文链接 if not self._is_paper_url(href): continue # 避免重复 if href in seen_urls: continue seen_urls.add(href) # 解析论文 paper = self._parse_paper_link(link, soup) if paper: # 应用年份过滤 if year_from or year_to: paper_year = paper.published_date.year if paper.published_date else None if paper_year: if year_from and paper_year < year_from: continue if year_to and paper_year > year_to: continue papers.append(paper) logger.info(f"RePEc search found {len(papers)} papers for query: {query}") except requests.Timeout: logger.error("RePEc search timed out") except requests.RequestException as e: logger.error(f"RePEc search request failed: {e}") except Exception as e: logger.error(f"RePEc search error: {e}") return papers[:max_results]
• paper_find_mcp/server.py:95-113 (helper)

  Generic _search helper function used by all platform-specific tools (including search_repec). Retrieves searcher instance from SEARCHERS dict, calls its .search() method, converts Paper objects to dicts for MCP response.
  async def _search( searcher_name: str, query: str, max_results: int = 10, **kwargs ) -> List[Dict]: """通用搜索函数""" searcher = SEARCHERS.get(searcher_name) if not searcher: logger.error(f"Unknown searcher: {searcher_name}") return [] try: papers = searcher.search(query, max_results=max_results, **kwargs) return [paper.to_dict() for paper in papers] except Exception as e: logger.error(f"Search failed for {searcher_name}: {e}") return []
• paper_find_mcp/server.py:75-85 (registration)

  Global SEARCHERS dictionary instantiation including 'repec': RePECSearcher(), providing the singleton searcher instance used by _search for the search_repec tool.
  SEARCHERS = { 'arxiv': ArxivSearcher(), 'pubmed': PubMedSearcher(), 'biorxiv': BioRxivSearcher(), 'medrxiv': MedRxivSearcher(), 'google_scholar': GoogleScholarSearcher(), 'iacr': IACRSearcher(), 'semantic': SemanticSearcher(), 'crossref': CrossRefSearcher(), 'repec': RePECSearcher(), }
• paper_find_mcp/server.py:732-816 (schema)

  Input schema defined by function parameters (query required, others optional with defaults) and comprehensive docstring describing usage, examples, and return format (List[Dict] with paper metadata).
  async def search_repec( query: str, max_results: int = 10, year_from: Optional[int] = None, year_to: Optional[int] = None, search_field: str = 'all', sort_by: str = 'relevance', doc_type: str = 'all', series: Optional[str] = None, ) -> List[Dict]: """Search economics papers on RePEc/IDEAS - the largest open economics bibliography. USE THIS TOOL WHEN: - Searching for ECONOMICS research (macro, micro, finance, econometrics) - You need working papers from NBER, Federal Reserve, World Bank, etc. - You want to find papers by JEL classification - Searching for economic policy analysis COVERAGE: 4.5M+ items including: - Working Papers: NBER, Fed banks, ECB, IMF, World Bank - Journal Articles: AER, JPE, QJE, Econometrica, etc. - Books and Book Chapters SEARCH SYNTAX: - Boolean: + for AND, | for OR, ~ for NOT (e.g., 'money ~liquidity') - Phrase: use double quotes (e.g., '"monetary policy"') - Author(Year): e.g., 'Acemoglu (2019)' or 'Kydland Prescott (1977)' - Synonyms: automatic (labor=labour, USA=United States) - Word stemming: automatic (find matches finds, finding, findings) LIMITATION: RePEc provides metadata only, not full PDFs. PDFs are hosted at original institutions (often freely available). Args: query: Search terms with optional boolean operators. max_results: Number of results (default: 10). year_from: Optional start year filter (e.g., 2020). year_to: Optional end year filter (e.g., 2025). search_field: Where to search, one of: - 'all': Whole record (default) - 'abstract': Abstract only - 'keywords': Keywords only - 'title': Title only - 'author': Author only sort_by: How to sort results, one of: - 'relevance': Most relevant (default) - 'newest': Most recent first - 'oldest': Oldest first - 'citations': Most cited first - 'recent_relevant': Recent and relevant - 'relevant_cited': Relevant and cited doc_type: Document type filter, one of: - 'all': All types (default) - 'articles': Journal articles - 'papers': Working papers (NBER, Fed, etc.) - 'chapters': Book chapters - 'books': Books - 'software': Software components series: Institution/journal series to search within, one of: - Institutions: 'nber', 'imf', 'worldbank', 'ecb', 'bis', 'cepr', 'iza' - Federal Reserve: 'fed', 'fed_ny', 'fed_chicago', 'fed_stlouis' - Top 5 Journals: 'aer', 'jpe', 'qje', 'econometrica', 'restud' - Other journals: 'jfe', 'jme', 'aej_macro', 'aej_micro', 'aej_applied' Returns: List of paper dicts with: paper_id (RePEc handle), title, authors, abstract, published_date, url, categories (JEL codes). Example: search_repec('inflation', series='nber') # Search NBER only search_repec('causal inference', series='aer', sort_by='newest') search_repec('machine learning', series='fed', year_from=2020) """ kwargs = { 'year_from': year_from, 'year_to': year_to, 'search_field': search_field, 'sort_by': sort_by, 'doc_type': doc_type, 'series': series, } # Remove None values kwargs = {k: v for k, v in kwargs.items() if v is not None} return await _search('repec', query, max_results, **kwargs)