Fix: Create FastAPI app first, mount Gradio onto it

Custom routes now take precedence over Gradio's catch-all.
Use gr.mount_gradio_app() to properly integrate both apps.

app.py

@@ -11,7 +11,7 @@ import re
 import sys
 import warnings
 
-from fastapi import Request
+from fastapi import FastAPI, Request
 from fastapi.responses import StreamingResponse
 
 from diffusers import AutoencoderKL, FlowMatchEulerDiscreteScheduler
@@ -500,7 +500,11 @@ init_app()
 pipe.transformer.layers._repeated_blocks = ["ZImageTransformerBlock"]
 spaces.aoti_blocks_load(pipe.transformer.layers, "zerogpu-aoti/Z-Image", variant="fa3")
 
-with gr.Blocks(title="Z-Image Generation MCP") as demo:
+css = """
+.fillable{max-width: 1230px !important}
+"""
+
+with gr.Blocks(title="Z-Image Generation MCP", css=css) as demo:
     gr.Markdown(
         """<div align="center">
 
@@ -586,11 +590,11 @@ with gr.Blocks(title="Z-Image Generation MCP") as demo:
         api_name="generate_image",  # nice, stable name for tool clients
     )
 
-# Get the underlying FastAPI app created by Gradio
-app = demo.app  # type: ignore[attr-defined]
+# ==================== FastAPI App with Custom Routes ====================
+# Create our own FastAPI app so custom routes take precedence over Gradio's catch-all
+app = FastAPI()
 
 
-# ==================== /generate/<prompt>.png Endpoint ====================
 @app.get("/generate/{prompt_path:path}")
 def http_generate_png(
     prompt_path: str,
@@ -691,10 +695,9 @@ def http_generate_png(
     return StreamingResponse(io.BytesIO(png_bytes), media_type="image/png")
 
 
-# =========================================================================
+# Mount Gradio app onto our FastAPI app
+app = gr.mount_gradio_app(app, demo, path="/")
 
-css = """
-.fillable{max-width: 1230px !important}
-"""
 if __name__ == "__main__":
-    demo.launch(css=css, mcp_server=True)
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=7860)