"""
Gradio + OpenAI MCP Connector — Clean, Fast, Streaming, With File Upload
"""

import os
import shutil
import gradio as gr
from openai import OpenAI

# ---------------------
# CONFIGURATION
# ---------------------
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
MCP_SERVER_URL = "https://mcp-1st-birthday-auto-deployer.hf.space/gradio_api/mcp/"

MODEL_FAST = "gpt-5-mini"   # for tool resolution / MCP calls
MODEL_STREAM = "gpt-5.1"       # for final streaming reply

client = OpenAI(api_key=OPENAI_API_KEY)

# ---------------------
# SYSTEM PROMPT
# ---------------------
SYSTEM_PROMPT = """
You are a fast MLOps automation assistant equipped with remote MCP tools
for dataset analysis, model training, evaluation, and deployment.

Rules:
- Use MCP tools when they directly address the user's request.
- Treat the uploaded CSV file URL as the source of truth. Never modify it.
- Never hallucinate tool names, arguments, or fields.
- Keep your internal reasoning hidden.
- Keep responses short, direct, and practical.

Workflow:
1) Decide if a tool is needed for the request.
2) If needed, call the correct MCP tool with the exact schema.
3) After tools complete, give a concise summary in plain language.
4) If no tool is needed, answer directly and briefly.
"""

# ---------------------
# NATIVE MCP CONNECTOR (HTTP STREAMING)
# ---------------------
TOOLS = [
    {
        "type": "mcp",
        "server_label": "deploy_tools",
        "server_url": MCP_SERVER_URL,  # HTTP streaming MCP server
    }
]


# ---------------------
# FILE UPLOAD HANDLER
# ---------------------
def handle_upload(file_obj, request: gr.Request):
    """
    - Persist uploaded file to a stable /tmp path
    - Return a public URL that the MCP tools can use directly
    """
    if file_obj is None:
        return None

    local_path = file_obj.name
    stable_path = os.path.join("/tmp", os.path.basename(local_path))

    try:
        shutil.copy(local_path, stable_path)
        local_path = stable_path
    except Exception:
        # If copy fails, still try original path
        pass

    base = str(request.base_url).rstrip("/")
    return f"{base}/gradio_api/file={local_path}"


# ---------------------
# MAIN CHAT HANDLER (STREAMING)
# ---------------------
def chat_send_stream(user_msg, history, file_url):
    """
    2-phase pipeline:
      PHASE 1: Non-streaming tool resolution using MODEL_FAST
      PHASE 2: Streaming final answer using MODEL_STREAM
    Gradio Chatbot expects: list[{"role": "...", "content": "..."}]
    """

    # Ensure history is list[dict(role, content)]
    if history is None:
        history = []

    # Append the user message to the UI history first
    history.append({"role": "user", "content": user_msg})

    # ---- Build messages for OpenAI (sanitize away metadata etc.) ----
    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
    for msg in history:
        role = msg.get("role")
        content = msg.get("content", "")
        if role in ("user", "assistant"):
            messages.append({"role": role, "content": content})

    # Inject file context into *last* user message for the model
    if file_url:
        last_user = messages[-1]
        if last_user["role"] == "user":