app.py

import os

import gradio as gr

from smolagents import CodeAgent, ToolCallingAgent, MCPClient, InferenceClientModel
from openai import OpenAI


model_name = None
workspace = "imessam"
environment = None
app_name = "example-vllm-openai-compatible"
function_name = "serve"

api_key = os.getenv("MODAL_API_KEY")

client = OpenAI(api_key=api_key)

prefix = workspace + (f"-{environment}" if environment else "")

client.base_url = (
    f"https://{prefix}--{app_name}-{function_name}.modal.run/v1"
)

print(str(client.base_url.host))

model = client.models.list().data[0]
model_id = model.id


def generate_podcast(prompt : str, history: list) -> str:

    response = ""

    try:
        mcp_client = MCPClient(
            {"url": "https://agents-mcp-hackathon-websearch.hf.space/gradio_api/mcp/sse", "transport": "sse"}# This is the MCP Server we created in the previous section
        )
        tools = mcp_client.get_tools()

        model = InferenceClientModel()
        agent = CodeAgent(tools=[*tools], model=model)

        response = str(agent.run(prompt))
        

    finally:
        mcp_client.disconnect()


    return response

demo = gr.ChatInterface(
    fn=generate_podcast,
    type="messages",
    examples=["Generate a podcast about AI"],
    title="Podcast Generator Agent and MCP Server",
    description="This is an agent that uses MCP tools to generate a podcast, and can be used as an MCP server.",
)

if __name__ == "__main__":
    demo.launch(
        mcp_server=True
    )