app.py

import os
import base64
import gradio as gr
from huggingface_hub import upload_file, InferenceClient
import json
from fastmcp import MCP, STIO

# --- Config ---
HF_DATASET_REPO = "OppaAI/Robot_MCP"
HF_VLM_MODEL = "Qwen/Qwen2.5-VL-7B-Instruct"

# --- MCP server instance ---
mcp = MCP()

# --- STIO for the LLM ---
stio = STIO(mcp)  # Bind STIO to MCP tools

# --- MCP Tool ---
@mcp.tools()
def say_hi(greeting_text: str = "Hi there!"):
    """Return a greeting command in JSON."""
    return {"command": "say_hi", "text": greeting_text}

# --- Helper Functions ---
def save_and_upload_image(image_b64, hf_token):
    image_bytes = base64.b64decode(image_b64)
    local_tmp_path = "/tmp/tmp.jpg"
    with open(local_tmp_path, "wb") as f:
        f.write(image_bytes)

    path_in_repo = "images/tmp.jpg"
    upload_file(
        path_or_fileobj=local_tmp_path,
        path_in_repo=path_in_repo,
        repo_id=HF_DATASET_REPO,
        token=hf_token,
        repo_type="dataset"
    )

    hf_image_url = f"https://huggingface.co/datasets/{HF_DATASET_REPO}/resolve/main/{path_in_repo}"
    return local_tmp_path, hf_image_url, path_in_repo, len(image_bytes)

# --- Main MCP function ---
def process_and_describe(payload: dict):
    try:
        hf_token = payload.get("hf_token")
        if not hf_token:
            return {"error": "HF token not provided in payload."}

        robot_id = payload.get("robot_id", "unknown")
        image_b64 = payload.get("image_b64")
        if not image_b64:
            return {"error": "No image provided."}

        # Save image & upload
        local_tmp_path, hf_url, path_in_repo, size_bytes = save_and_upload_image(image_b64, hf_token)

        # Initialize HF client
        hf_client = InferenceClient(token=hf_token)

        # --- System prompt with STIO instructions ---
        system_prompt = f"""
        You are a helpful robot assistant. You have access to MCP tools via STIO.
        When you receive an image, you must:
        1️⃣ Describe the image in detail.
        2️⃣ Decide actions for the robot. Example:
           - Human figure → call `say_hi` tool with a friendly greeting (vary every time)
        3️⃣ Use STIO to call the tools. Always respond in JSON if calling tools.

        Available tools:
        {stio.describe_tools()}
        """

        messages_payload = [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content": [
                {"type": "text", "text": "Here is an image."},
                {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{image_b64}"}}
            ]}
        ]

        # --- Call VLM with STIO ---
        chat_completion = hf_client.chat.completions.create(
            model=HF_VLM_MODEL,
            messages=messages_payload,
            max_tokens=300
        )

        vlm_text = chat_completion.choices[0].message.content.strip()

        # --- Use STIO to execute tool calls if present ---
        tool_results = stio.run(vlm_text)

        return {
            "saved_to_hf_hub": True,
            "repo_id": HF_DATASET_REPO,
            "path_in_repo": path_in_repo,
            "image_url": hf_url,
            "file_size_bytes": size_bytes,
            "robot_id": robot_id,
            "vlm_response": vlm_text,
            "tool_results": tool_results
        }

    except Exception as e:
        return {"error": f"An API error occurred: {str(e)}"}

# --- Gradio MCP Interface ---
demo = gr.Interface(
    fn=process_and_describe,
    inputs=gr.JSON(label="Input Payload"),
    outputs=gr.JSON(label="Reply to Jetson"),
    api_name="predict"
)

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