app.py

import os
import base64
import time
import io
import gradio as gr
from fastmcp import Client
from fastmcp.client import StreamableHttpTransport
import asyncio
from dotenv import load_dotenv
import ast  # For safely evaluating Python literals returned from server

# -------------------------------
# Load environment variables
# -------------------------------
load_dotenv()

ROBOT_ID = "Robot_MCP_Client"  # Local client identifier
HF_TOKEN = os.environ.get("HF_TOKEN")
if not HF_TOKEN:
    print("Warning: HF_TOKEN not found. API calls may fail.")
    HF_TOKEN = "missing_token_placeholder"  # Placeholder to avoid crash

# MCP server info
MCP_SERVER_URL = "https://oppaai-robot-mcp-server.hf.space/gradio_api/mcp/"
SERVER_NAME = "Robot_MCP_Server"
TOOL_NAME = "Robot_MCP_Server_robot_watch"

# -------------------------------
# Initialize MCP client globally
# -------------------------------
HTTP_TRANSPORT = StreamableHttpTransport(url=MCP_SERVER_URL)
MCP_CLIENT = Client(transport=HTTP_TRANSPORT, name=SERVER_NAME)


async def process_webcam_stream_async(image):
    """
    Send webcam image to MCP server and process the response.

    Args:
        image (PIL.Image or None): Image captured from webcam or uploaded.

    Returns:
        tuple: (description, human, objects_str, environment)
            description (str): Description of scene.
            human (str): Human-related information.
            objects_str (str): Comma-separated list of objects.
            environment (str): Environment description.
    """
    if image is None:
        return "", "", "", ""

    if HF_TOKEN == "missing_token_placeholder":
        return "Error: HF_TOKEN not set locally.", "", "", ""

    # Convert image to Base64 string
    buffered = io.BytesIO()
    image.save(buffered, format="JPEG")
    b64_img = base64.b64encode(buffered.getvalue()).decode("utf-8")

    # Prepare payload according to server's expected fields
    payload = {
        "hf_token_input": HF_TOKEN,
        "robot_id_input": ROBOT_ID,
        "image_b64_input": b64_img
    }

    try:
        # Use async context to call MCP server tool
        async with MCP_CLIENT:
            response = await MCP_CLIENT.call_tool(TOOL_NAME, payload)

            if response.is_error:
                # Extract error message
                error_text = response.content[0].text if response.content and isinstance(response.content, list) else "Unknown error"
                raise Exception(f"MCP Tool Error: {error_text}")

            # Server may return Python-style string (single quotes)
            raw_text = response.content[0].text
            response_dict = ast.literal_eval(raw_text)

            # -------------------------------
            # Extract fields from response
            # -------------------------------
            vlm_result = response_dict.get("result", {})

            description_out = vlm_result.get("description", "")
            human_out = vlm_result.get("human", "")
            objects_list = vlm_result.get("objects", [])
            environment_out = vlm_result.get("environment", "")

            # Convert objects list to a comma-separated string for display
            objects_str = ", ".join(objects_list) if isinstance(objects_list, list) else str(objects_list)

            return description_out, human_out, objects_str, environment_out

    except Exception as e:
        print(f"Error calling remote MCP API: {e}")
        import traceback
        traceback.print_exc()
        return f"Error: {e}", "", "", ""


# -------------------------------
# Gradio UI
# -------------------------------
with gr.Blocks() as demo:
    gr.Markdown("## 🎥 Robot Vision Webcam Stream (using MCP Client)")
    gr.Markdown("""   
    ### 🔑 Hugging Face Token Required
    To use this application, you must set a valid **Hugging Face API Token** in your local environment variables: `HF_TOKEN`.
    **A write token is required** to upload images to the public dataset associated with this space.
    Resource usage for VLM inference will be tracked against your account.
    """)
    with gr.Row():
        # Webcam / upload image input
        webcam_input = gr.Image(
            label="Captured from Web-Cam",
            sources=["upload", "webcam"],
            type="pil"
        )
        with gr.Column():
            # Output fields for MCP response
            description_out = gr.Textbox(label="Description", lines=5)
            human_out = gr.Textbox(label="Human", lines=3)
            objects_out = gr.Textbox(label="Objects", lines=2)
            environment_out = gr.Textbox(label="Environment", lines=3)

    # Stream webcam input to server every 0.5 seconds
    webcam_input.stream(
        process_webcam_stream_async,
        inputs=[webcam_input],
        outputs=[description_out, human_out, objects_out, environment_out],
        stream_every=0.5
    )

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