Update server.py

server.py

@@ -5,6 +5,7 @@ from llama_index.core import (
     StorageContext,
     load_index_from_storage,
 )
+from llama_index.tools.arxiv import ArxivToolSpec
 from llama_index.core import Settings
 from llama_index.llms.azure_openai import AzureOpenAI
 from llama_index.embeddings.azure_openai import AzureOpenAIEmbedding
@@ -94,6 +95,75 @@ port = int(os.getenv("PORT", 7860))
 mcp = FastMCP("OnBase", port=port)
 
 
+# Instancia global de ArXiv
+paper_tool = ArxivToolSpec()
+
+@mcp.tool()
+async def search_arxiv_papers(
+    ctx: Context,
+    query: str,
+    max_results: Optional[int] = 5,
+    sort_by: Optional[str] = "relevance"
+) -> dict:
+    """
+    Search for academic papers on ArXiv using natural language queries.
+    
+    Args:
+        query: Natural language search query (e.g. "machine learning in healthcare")
+        max_results: Maximum number of results to return (default 5, max 10)
+        sort_by: Sorting method ("relevance" or "last_updated_date")
+    
+    Returns:
+        dict: {
+            "papers": List of paper summaries,
+            "count": Number of results,
+            "query": Original query,
+            "status": "success" or "error"
+        }
+    """
+    try:
+        # Validar parámetros
+        max_results = min(max(1, max_results), 10)  # Limitar entre 1 y 10
+        if sort_by not in ["relevance", "last_updated_date"]:
+            sort_by = "relevance"
+        
+        # Usar el tool de ArXiv
+        results = paper_tool.arxiv_search(
+            query=query,
+            max_results=max_results,
+            sort_by=sort_by
+        )
+        
+        # Procesar resultados
+        papers = []
+        for paper in results:
+            papers.append({
+                "title": paper.metadata.get("Title", ""),
+                "authors": paper.metadata.get("Authors", ""),
+                "abstract": paper.metadata.get("Summary", ""),
+                "published": paper.metadata.get("Published", ""),
+                "pdf_url": paper.metadata.get("PDF url", ""),
+                "arxiv_id": paper.metadata.get("Entry ID", "").split('/')[-1]
+            })
+        
+        return {
+            "papers": papers,
+            "count": len(papers),
+            "query": query,
+            "status": "success"
+        }
+        
+    except Exception as e:
+        await ctx.error(f"Error in ArXiv search: {str(e)}")
+        return {
+            "papers": [],
+            "count": 0,
+            "query": query,
+            "status": "error",
+            "error": str(e)
+        }
+
+
 
 @mcp.tool()
 async def list_retrievers(source: str = None) -> dict: