main.py

import os
import pprint
import uuid
from dotenv import load_dotenv

from langchain_mcp_adapters.client import MultiServerMCPClient
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import ToolNode
from langgraph.graph import MessagesState, END, StateGraph
from langchain_core.messages import HumanMessage
from langgraph.checkpoint.memory import MemorySaver


from pmcp.agents.executor import ExecutorAgent
from pmcp.agents.trello_agent import TrelloAgent
from pmcp.agents.github_agent import GithubAgent
from pmcp.agents.planner import PlannerAgent

from pmcp.nodes.human_interrupt_node import HumanInterruptNode
from pmcp.nodes.human_resume_node import HumanResumeNode

from pmcp.models.state import PlanningState

load_dotenv()


async def call_llm(llm_with_tools: ChatOpenAI, state: MessagesState):
    response = llm_with_tools.invoke(state["messages"])
    return {"messages": [response]}


async def main():
    mcp_client_trello = MultiServerMCPClient(
        {
            "trello": {
                "command": "python",
                "args": [os.getenv("MCP_TRELLO_PATH")],
                "transport": "stdio",
            }
        }
    )
    mcp_client_github = MultiServerMCPClient(
        {
            "github": {
                "command": "python",
                "args": [os.getenv("MCP_GITHUB_PATH")],
                "transport": "stdio",
            }
        }
    )

    memory = MemorySaver()

    trello_tools = await mcp_client_trello.get_tools()
    github_tools = await mcp_client_github.get_tools()

    tool_node = ToolNode(github_tools + trello_tools)

    llm = ChatOpenAI(
        model="Qwen/Qwen2.5-32B-Instruct",
        temperature=0.0,
        api_key=os.getenv("NEBIUS_API_KEY"),
        base_url="https://api.studio.nebius.com/v1/",
    )

    trello_agent = TrelloAgent(
        tools=trello_tools,
        llm=llm,
    )

    github_agent = GithubAgent(llm=llm, tools=github_tools)

    planner_agent = PlannerAgent(
        llm=llm,
    )
    executor_agent = ExecutorAgent(llm=llm)

    human_interrupt_node = HumanInterruptNode(
        llm=llm,
    )
    human_resume_node = HumanResumeNode(llm=llm)

    graph = StateGraph(MessagesState)
    graph.add_node(planner_agent.agent.agent_name, planner_agent.acall_planner_agent)
    graph.add_node(trello_agent.agent.agent_name, trello_agent.acall_trello_agent)
    graph.add_node(github_agent.agent.agent_name, github_agent.acall_github_agent)
    graph.add_node(executor_agent.agent.agent_name, executor_agent.acall_executor_agent)
    graph.add_node("tool", tool_node)
    graph.add_node("human_interrupt", human_interrupt_node.call_human_interrupt_agent)
    graph.set_entry_point(planner_agent.agent.agent_name)

    def should_continue(state: PlanningState):
        last_message = state.messages[-1]
        if last_message.tool_calls:
            return "human_interrupt"
        return executor_agent.agent.agent_name

    def execute_agent(state: PlanningState):
        if state.current_step:
            return state.current_step.agent

        return END

    graph.add_conditional_edges(trello_agent.agent.agent_name, should_continue)
    graph.add_conditional_edges(github_agent.agent.agent_name, should_continue)
    graph.add_conditional_edges(executor_agent.agent.agent_name, execute_agent)

    graph.add_edge("tool", trello_agent.agent.agent_name)
    graph.add_edge("tool", github_agent.agent.agent_name)
    graph.add_edge(planner_agent.agent.agent_name, executor_agent.agent.agent_name)

    app = graph.compile(checkpointer=memory)
    app.get_graph(xray=True).draw_mermaid()

    user_input = input("user >")
    config = {
        "configurable": {"thread_id": f"{str(uuid.uuid4())}"},
        "recursion_limit": 100,
    }

    is_message_command = False
    while user_input.lower() != "q":

        if is_message_command:
            app_input = human_resume_node.call_human_interrupt_agent(
                user_input
            )
            is_message_command = False
        else:
            app_input = {
                "messages": [
                    HumanMessage(content=user_input),
                ]
            }
        async for res in app.astream(
            app_input,
            config=config,
            stream_mode="values",
        ):
            if "messages" in res:
                pprint.pprint(res["messages"][-1])
            else:
                pprint.pprint(res["__interrupt__"][0])
                is_message_command = True
            pprint.pprint("-------------------------------------")
        user_input = input("user >")


if __name__ == "__main__":
    import asyncio

    asyncio.run(main())