feat: Migrate agent to streamable HTTP, streamline app with environment variable configuration and a unified chat/file upload UI, and add dev utility files.

.gitignore

@@ -0,0 +1,2 @@
+.venv
+__pycache__

agent.py

@@ -1,7 +1,7 @@
 import os
 import asyncio
 from typing import Any, Optional
-from mcp.client.sse import sse_client
+from mcp.client.streamable_http import streamablehttp_client
 from mcp.client.session import ClientSession
 from agents import Agent, Runner
 import openai
@@ -9,6 +9,7 @@ import openai
 # Define the MCP Server URL
 MCP_SERVER_URL = "https://mcp-1st-birthday-auto-deployer.hf.space/gradio_api/mcp/"
 
+class MCPAgent:
     def __init__(self, provider: str, api_key: str, mcp_server_url: str, base_url: Optional[str] = None):
         self.provider = provider
         self.api_key = api_key
@@ -20,10 +21,10 @@ MCP_SERVER_URL = "https://mcp-1st-birthday-auto-deployer.hf.space/gradio_api/mcp
 
     async def initialize(self):
         """Connects to the MCP server and initializes the OpenAI Agent."""
-        # 1. Connect to MCP Server via SSE
-        self.sse_context = sse_client(self.mcp_server_url)
+        # 1. Connect to MCP Server via Streamable HTTP
+        self.sse_context = streamablehttp_client(self.mcp_server_url)
         self.transport = await self.sse_context.__aenter__()
-        self.session = ClientSession(self.transport, self.transport)
+        self.session = ClientSession(self.transport[0], self.transport[1])
         await self.session.__aenter__()
         await self.session.initialize()
 
@@ -60,7 +61,7 @@ MCP_SERVER_URL = "https://mcp-1st-birthday-auto-deployer.hf.space/gradio_api/mcp
             name="MLOps Agent",
             instructions="You are an expert MLOps agent. You have access to tools for analyzing data, training models, and deploying them. Use them to help the user.",
             tools=agent_tools,
-            model="gpt-4o", # Default, can be overridden if needed
+            model="gpt-5-mini-2025-08-07", # Default, can be overridden if needed
         )
 
     async def chat(self, user_message: str):

app.py

@@ -2,101 +2,79 @@ import gradio as gr
 import asyncio
 from agent import MCPAgent
 import os
+from dotenv import load_dotenv
 
-# Global variable to store agent instances per session (simplified for demo)
-# In a real production app, you'd want more robust session management.
+load_dotenv()
+
+# Configuration from Environment Variables
+OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
+MCP_SERVER_URL = os.getenv("MCP_SERVER_URL", "https://mcp-1st-birthday-auto-deployer.hf.space/gradio_api/mcp/")
+PROVIDER = "OpenAI" # Default to OpenAI
+
+# Global variable to store agent instances per session
 agents = {}
 
-async def get_or_create_agent(session_id, provider, api_key, mcp_server_url, base_url):
+async def get_or_create_agent(session_id):
     if session_id not in agents:
-        agent = MCPAgent(provider, api_key, mcp_server_url, base_url)
+        if not OPENAI_API_KEY:
+            raise gr.Error("OPENAI_API_KEY environment variable is not set.")
+        
+        agent = MCPAgent(PROVIDER, OPENAI_API_KEY, MCP_SERVER_URL)
         await agent.initialize()
         agents[session_id] = agent
     return agents[session_id]
 
-async def chat_fn(message, history, provider, api_key, mcp_server_url, base_url, request: gr.Request):
+async def chat_fn(message, history, file_obj, request: gr.Request):
     if not message:
         return history
     
-    if not api_key:
-        raise gr.Error("Please enter an API Key.")
-
-    session_id = request.session_hash
-    try:
-        agent = await get_or_create_agent(session_id, provider, api_key, mcp_server_url, base_url)
-        response = await agent.chat(message)
-        history.append((message, str(response)))
+    if not file_obj:
+        # Enforce file upload
+        history.append({"role": "user", "content": message})
+        history.append({"role": "assistant", "content": "Please upload a CSV file first."})
         return history, ""
-    except Exception as e:
-        raise gr.Error(f"Error: {str(e)}")
 
-async def one_shot_fn(file_obj, provider, api_key, mcp_server_url, base_url, request: gr.Request):
-    if not api_key:
-        raise gr.Error("Please enter an API Key.")
-    
     session_id = request.session_hash
     try:
-        agent = await get_or_create_agent(session_id, provider, api_key, mcp_server_url, base_url)
+        agent = await get_or_create_agent(session_id)
+        
+        # Pass file path context if needed, or just assume agent knows tools
+        # For now, we append the file path to the message if it's the first interaction or just let the agent handle it via tools.
+        # The user said "ask user to load the CSV file first and then let it type the question".
+        # We can prepend the file info to the message invisibly or just rely on the agent having access to the file via a tool if we uploaded it.
+        # But wait, the agent needs to know WHERE the file is.
+        # The previous `one_shot_fn` implied using `upload_file_to_gradio` tool.
+        # Since we are in a chat loop, we might need to tell the agent about the file once.
+        
+        # Let's just pass the message as is, but maybe prepend context about the file if it's new?
+        # Or better: The agent has tools. One tool is `Auto_Deployer_web_analyze`.
+        # It takes a file path.
+        # We should probably tell the agent: "I have uploaded a file at {file_obj.name}. {message}"
         
-        # Construct the prompt
-        prompt = "Please perform the full MLOps pipeline: Analyze the data, train a model, deploy it, test the API, and provide a final report."
-        if file_obj:
-            # In a real scenario, we would upload this file using the web_auto_deploy tool
-            # For now, we pass the path if the agent can handle local paths, or we need to upload it.
-            # The prompt implies the agent uses tools.
-            prompt += f" Use this file: {file_obj.name}"
+        full_message = f"I have uploaded a file at {file_obj.name}. {message}"
         
-        response = await agent.chat(prompt)
-        return str(response)
+        response = await agent.chat(full_message)
+        
+        history.append({"role": "user", "content": message})
+        history.append({"role": "assistant", "content": str(response)})
+        return history, ""
     except Exception as e:
-        return f"Error: {str(e)}"
+        history.append({"role": "user", "content": message})
+        history.append({"role": "assistant", "content": f"Error: {str(e)}"})
+        return history, ""
 
 with gr.Blocks(title="MCP MLOps Agent") as demo:
     gr.Markdown("# 🤖 MCP MLOps Agent")
     gr.Markdown("Powered by OpenAI Agents SDK & MCP 1st Birthday Hackathon Tools")
     
-    with gr.Sidebar():
-        provider = gr.Dropdown(
-            choices=["OpenAI", "OpenAI Compatible"], 
-            value="OpenAI", 
-            label="Provider"
-        )
-        api_key = gr.Textbox(
-            label="API Key", 
-            type="password",
-            placeholder="sk-..."
-        )
-        mcp_server_url = gr.Textbox(
-            label="MCP Server URL",
-            value="https://mcp-1st-birthday-auto-deployer.hf.space/gradio_api/mcp/",
-            placeholder="https://..."
-        )
-        base_url = gr.Textbox(
-            label="Base URL (for Compatible Providers)", 
-            placeholder="https://api.nebius.ai/v1",
-            visible=False
-        )
-        
-        def update_base_url(p):
-            return gr.update(visible=(p == "OpenAI Compatible"))
-        
-        provider.change(update_base_url, inputs=provider, outputs=base_url)
-
-    with gr.Tabs():
-        with gr.Tab("💬 Chat"):
-            chatbot = gr.Chatbot(height=600)
-            msg = gr.Textbox(label="Message")
-            clear = gr.ClearButton([msg, chatbot])
-            
-            msg.submit(chat_fn, [msg, chatbot, provider, api_key, mcp_server_url, base_url], [chatbot, msg])
-
-        with gr.Tab("🚀 One-Shot Workflow"):
-            gr.Markdown("Upload a dataset and let the agent handle the rest: Analyze -> Train -> Deploy -> Test -> Report")
-            file_input = gr.File(label="Upload Dataset (CSV)")
-            run_btn = gr.Button("Run Full Pipeline", variant="primary")
-            output_md = gr.Markdown(label="Agent Report")
-            
-            run_btn.click(one_shot_fn, [file_input, provider, api_key, mcp_server_url, base_url], [output_md])
+    with gr.Row():
+        file_input = gr.File(label="Upload Dataset (CSV)")
+    
+    chatbot = gr.Chatbot(height=600, type="messages", allow_tags=False)
+    msg = gr.Textbox(label="Message", placeholder="Type your message here...")
+    clear = gr.ClearButton([msg, chatbot, file_input])
+    
+    msg.submit(chat_fn, [msg, chatbot, file_input], [chatbot, msg])
 
 if __name__ == "__main__":
     demo.queue().launch()

requirements.txt

@@ -1,4 +1,4 @@
-gradio
-openai
-mcp
-agents-sdk
+gradio==5.50.0
+openai==2.8.1
+mcp==1.10.1
+openai-agents==0.1.0

test_warning.py

@@ -0,0 +1,8 @@
+import gradio as gr
+
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot(type="messages", allow_tags=False)
+
+if __name__ == "__main__":
+    print("Running demo...")
+    # demo.launch()