conductor_server.py

"""
MCP Server exposing the Conductor Agent functionality.
"""

import asyncio
from agents import Agent, Runner
from mcp.server.fastmcp import FastMCP

# Import the sub-agents used by the Conductor
from search_agent import search_agent
from classify_agent import classify_agent
from compose_agent import compose_agent

# Define the Conductor Agent (originally test_agent in main.py)
# This agent uses the other agents as tools to perform the orchestration.
conductor_agent = Agent(
    name="Conductor",
    instructions="You are the mighty orchestrator. Your workflow is as follows: "
    "1. Use the search tool to find the list of events in the city provided by the user input. "
    "2. For each event found, use the classifier tool to determine the most relevant group of people interested in it. "
    "3. For each event and its corresponding audience group, use the compose tool to generate a push notification. "
    "4. Format the final output for EACH processed event as follows: "
    "   Line 1: The event name"
    "   Line 2: The determined audience group name"
    "   Line 3: The generated push notification text"
    "Separate the entries for each event with a single empty line.",
    tools=[
        search_agent.as_tool(tool_name="search", tool_description="Searches for events in a given city and returns a list of event details."),
        classify_agent.as_tool(tool_name="classifier", tool_description="Takes event details (and potentially user PII from the prompt context) and determines the target audience groups."),
        compose_agent.as_tool(tool_name="compose", tool_description="Generates a push notification text for a given event name and target audience group.")
    ],
    model="gpt-4.1", # Ensure consistency or use appropriate model
)

# Initialize the MCP Server
mcp = FastMCP("Conductor Event Pipeline Server")


@mcp.tool(description="Runs a complete pipeline to find events in a specified city, "
          "classify their target audience, and generate tailored push notifications. "
          "Use this when a user wants to discover events and get notifications ready "
          "for different groups.")
async def run_event_pipeline(city: str) -> str:
    """
    Runs the full event processing pipeline for a given city.

    1. Searches for events in the city.
    2. Classifies the audience for each event.
    3. Composes a push notification for each event and audience pair.

    Args:
        city: The name of the city to search for events in.

    Returns:
        A string containing the formatted results (event name, group, notification)
        for each event, separated by empty lines.
    """
    print(f"Received request to run event pipeline for city: {city}") # Server-side log
    try:
        # Use the Runner to execute the conductor_agent with the city name as input
        # The agent's instructions will guide the LLM to use the configured tools.
        result = await Runner.run(conductor_agent, city)
        print(f"Pipeline result: {result.final_output}") # Server-side log
        return result.final_output
    except Exception as e:
        print(f"Error running event pipeline for city {city}: {e}")
        # Depending on requirements, might return an error message or raise exception
        return f"An error occurred while processing events for {city}: {str(e)}"


if __name__ == "__main__":
    # Run the MCP server (likely via stdio by default with mcp-sdk)
    mcp.run()