Auto-start FastAPI and MCP server for Hugging Face Spaces deployment

app.py

@@ -2728,11 +2728,68 @@ with gr.Blocks(
 
 if __name__ == "__main__":
     import os
+    import threading
+    import time
+    import requests
+    
     # For Hugging Face Spaces, bind to 0.0.0.0; for local dev, use 127.0.0.1
     # HF Spaces sets SPACE_ID environment variable
     is_hf_space = os.getenv("SPACE_ID") is not None
     server_name = "0.0.0.0" if is_hf_space else "127.0.0.1"
     
+    # Start backend services if running in HF Spaces
+    if is_hf_space:
+        def start_mcp_server():
+            """Start MCP server in background thread."""
+            try:
+                import uvicorn
+                from backend.mcp_server import server
+                uvicorn.run(
+                    server.app,
+                    host="0.0.0.0",
+                    port=int(os.getenv("MCP_PORT", "8900")),
+                    log_level="info"
+                )
+            except Exception as e:
+                print(f"Warning: Could not start MCP server: {e}")
+        
+        def start_fastapi_server():
+            """Start FastAPI server in background thread."""
+            try:
+                import uvicorn
+                uvicorn.run(
+                    "backend.api.main:app",
+                    host="0.0.0.0",
+                    port=int(os.getenv("API_PORT", "8000")),
+                    log_level="info"
+                )
+            except Exception as e:
+                print(f"Warning: Could not start FastAPI server: {e}")
+        
+        # Start services in background threads
+        print("Starting backend services...")
+        mcp_thread = threading.Thread(target=start_mcp_server, daemon=True)
+        api_thread = threading.Thread(target=start_fastapi_server, daemon=True)
+        
+        mcp_thread.start()
+        time.sleep(2)  # Give MCP server time to start
+        api_thread.start()
+        time.sleep(3)  # Give FastAPI time to start
+        
+        # Wait for services to be ready
+        max_attempts = 30
+        for attempt in range(max_attempts):
+            try:
+                response = requests.get("http://localhost:8000/health", timeout=2)
+                if response.status_code == 200:
+                    print("✅ FastAPI backend is ready!")
+                    break
+            except:
+                if attempt < max_attempts - 1:
+                    time.sleep(1)
+                else:
+                    print("⚠️ FastAPI backend may not be ready, continuing anyway...")
+    
     demo.launch(
         server_name=server_name,
         server_port=7860,