使用自訂義的GPTs

youtube 總結

backend.py

import os
import glob
import uuid
import shutil
import yt_dlp
import re
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel

app = FastAPI(
    title="yt-mcp-server",
    servers=[
        {
            "url": "https://information-stationery-nations-conventional.trycloudflare.com",  # 例如 ngrok / cloudflare tunnel 網址
            "description": "Public HTTPS endpoint"
        }
    ],
)

class VideoRequest(BaseModel):
    url: str
    lang: str | None = None

def parse_vtt(vtt_content):
    lines = vtt_content.split('\n')
    cues = []
    
    # Regex for timestamp: 00:00:22.633 or 00:22.633
    time_pattern = re.compile(r'(\d{2}:)?(\d{2}):(\d{2})\.(\d{3})')
    
    current_start = None
    current_lines = []
    
    for line in lines:
        line = line.strip()
        if '-->' in line:
            # Save previous cue if exists
            if current_start is not None and current_lines:
                cues.append({'start': current_start, 'text': ' '.join(current_lines)})
            
            # New cue
            parts = line.split(' --> ')
            start_str = parts[0]
            match = time_pattern.match(start_str)
            if match:
                groups = match.groups()
                h = int(groups[0].replace(':', '')) if groups[0] else 0
                m = int(groups[1])
                s = int(groups[2])
                current_start = h * 3600 + m * 60 + s
                current_lines = []
            else:
                current_start = None 
        elif line and not line.isdigit() and not line.startswith('WEBVTT') and not line.startswith('Kind:') and not line.startswith('Language:') and not line.startswith('NOTE'):
             clean = re.sub(r'<[^>]+>', '', line)
             current_lines.append(clean)
             
    # Append last cue
    if current_start is not None and current_lines:
        cues.append({'start': current_start, 'text': ' '.join(current_lines)})
        
    # Merge cues close in time (e.g. within 5 seconds)
    merged = {}
    if not cues:
        return merged
        
    last_key = cues[0]['start']
    merged[last_key] = cues[0]['text']
    
    for cue in cues[1:]:
        if cue['start'] - last_key < 5:
            merged[last_key] += ' ' + cue['text']
        else:
            last_key = cue['start']
            merged[last_key] = cue['text']
            
    return merged

@app.post("/yt")
def get_subtitles(request: VideoRequest):
    """
    Accepts a YouTube URL and returns the subtitles.
    """
    print(f"Processing request for URL: {request.url} with lang: {request.lang}")
    url = request.url
    lang = request.lang
    
    # Create a unique temporary directory for this request to avoid collisions
    request_id = str(uuid.uuid4())
    temp_dir = f"/tmp/{request_id}"
    os.makedirs(temp_dir, exist_ok=True)
    
    ydl_opts = {
        'skip_download': True,
        'writesubtitles': True,
        'writeautomaticsub': True,
        'outtmpl': f'{temp_dir}/%(id)s',
        'quiet': True,
        'no_warnings': True,
        'http_headers': {
            'User-Agent': 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36',
        },
    }
    if lang:
        ydl_opts['subtitleslangs'] = [lang]
    else:
        # If no language specified, limit to common languages to avoid 429 (Too Many Requests)
        # from fetching all available auto-generated captions.
        ydl_opts['subtitleslangs'] = ['en', 'zh-Hant', 'zh-TW', 'ja']

    try:
        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
            try:
                info = ydl.extract_info(url, download=True)
            except Exception as e:
                raise HTTPException(status_code=400, detail=f"Failed to fetch video info: {str(e)}")

            video_id = info.get('id')
            title = info.get('title', 'Untitled Video')
            
            # Find the subtitle file
            # Since we use a unique temp dir, we can just look for any file starting with video_id
            # This handles cases where requested 'zh' results in 'zh-TW' file
            pattern = f"{temp_dir}/{video_id}.*"
            files = glob.glob(pattern)
            
            if not files:
                raise HTTPException(status_code=404, detail=f"No subtitles found.")
            
            # Use the first matching file (usually .vtt)
            subtitle_file = files[0]
            
            with open(subtitle_file, 'r', encoding='utf-8') as f:
                content = f.read()
            
            transcribed_part = parse_vtt(content)

            return {
                "video_id": video_id,
                "title": title,
                "page": 1,
                "total_pages": 1,
                "transcribed_part": transcribed_part
            }

    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Internal server error: {str(e)}")
    finally:
        # Clean up temporary directory
        if os.path.exists(temp_dir):
            shutil.rmtree(temp_dir)

if __name__ == "__main__":
    import uvicorn
    # Run the server
    uvicorn.run(app, host="0.0.0.0", port=8000)

public domain exposure

cloudflared tunnel --url http://127.0.0.1:8000

建立動作

從 URL 匯入 e.g. https://information-stationery-nations-conventional.trycloudflare.com/openapi.json

撰寫 prompt 並測試