arXiv MCP Server

"""arXiv MCP Server - Search and read arXiv papers.""" import os from pathlib import Path from mcp.server.fastmcp import FastMCP from .arxiv_client import search_papers, download_pdf, Paper from .pdf_parser import extract_text, extract_section from .storage import PaperStorage # Initialize storage STORAGE_DIR = Path(os.environ.get("ARXIV_STORAGE_DIR", Path.home() / ".arxiv-mcp" / "papers")) storage = PaperStorage(STORAGE_DIR) # Create MCP server mcp = FastMCP("arxiv-mcp-server") @mcp.tool() def search(query: str, max_results: int = 10) -> str: """Search arXiv papers by title, keywords, or arXiv ID. Args: query: Search query (title, keywords, or arXiv ID like 2401.12345) max_results: Maximum number of results to return (default: 10) Returns: List of matching papers with ID, title, authors, and abstract """ papers = search_papers(query, max_results) if not papers: return "No papers found." results = [] for p in papers: authors = ", ".join(p.authors[:3]) if len(p.authors) > 3: authors += " et al." results.append( f"**{p.id}**: {p.title}\n" f"Authors: {authors}\n" f"Published: {p.published}\n" f"Abstract: {p.abstract[:300]}...\n" ) return "\n---\n".join(results) @mcp.tool() def get_paper(paper_id: str, section: str = "all") -> str: """Get the full text of an arXiv paper. Args: paper_id: arXiv paper ID (e.g., "2401.12345") section: Which section to return: "all", "abstract", "introduction", "method", "conclusion" Returns: The paper text (full or specified section) """ # Check cache first cached_text = storage.get_text(paper_id) if cached_text: if section == "all": return cached_text return extract_section(cached_text, section) # Search for the paper papers = search_papers(paper_id, max_results=1) if not papers: return f"Paper {paper_id} not found." paper = papers[0] # Download PDF pdf_path = storage.get_pdf_path(paper_id) try: download_pdf(paper, str(pdf_path)) except Exception as e: return f"Failed to download paper: {e}" # Extract text try: full_text = extract_text(str(pdf_path)) except Exception as e: return f"Failed to extract text from PDF: {e}" # Cache results storage.save_paper_metadata(paper) storage.save_text(paper_id, full_text) if section == "all": return full_text return extract_section(full_text, section) @mcp.tool() def list_downloaded_papers() -> str: """List all locally downloaded papers. Returns: List of downloaded papers with their metadata """ papers = storage.list_papers() if not papers: return "No papers downloaded yet." results = [] for p in papers: authors = ", ".join(p["authors"][:3]) if len(p["authors"]) > 3: authors += " et al." results.append(f"**{p['id']}**: {p['title']}\nAuthors: {authors}") return "\n---\n".join(results) def main(): """Run the arXiv MCP server.""" mcp.run() if __name__ == "__main__": main()