yt-network-scraper

A network-first YouTube video scraper that extracts metadata, transcripts, comments, and summaries from captured network payloads using headless Chrome and Selenium.

Unlike traditional DOM scrapers, this package opens a real browser via Selenium, captures network responses through Chrome DevTools performance logs, and parses YouTube's own JSON payloads (ytInitialPlayerResponse, ytInitialData, ytcfg). It then uses YouTube's innertube API for transcripts and comments. This approach is more resilient to UI changes and avoids brittle CSS selectors.

Why yt-network-scraper?

There are several excellent YouTube libraries on PyPI. Here is how yt-network-scraper compares to the most popular ones, so you can pick the right tool for your use case:

When to use yt-network-scraper

Use this package if you need:

• Comments — very few libraries extract YouTube comments. This one does, with full author info, likes, hearted/pinned status, and reply counts.
• Dislike counts — integrated with the Return YouTube Dislike API, which no other listed library provides.
• Automatic summaries — built-in extractive summarization of transcripts, so you get a quick text summary alongside the full transcript.
• Access-block detection — detects CAPTCHAs, consent walls, and sign-in challenges and reports them rather than silently failing or trying to bypass them.
• A real browser session — some videos require JavaScript rendering and network-level payload capture that pure HTTP libraries cannot access. This package captures Chrome DevTools performance logs to extract YouTube's own JSON payloads.
• A single, unified result object — metadata, transcript, comments, engagement, summary, and network diagnostics in one typed VideoResult with to_dict() for JSON serialization.

Use a different package if you need:

• Video downloading → use yt-dlp or pytube
• Search or channel browsing → use ytscrape or tubescrape
• Transcripts only (lightweight, no browser) → use youtube-transcript-api
• Async / high-throughput scraping → use ytscrape or tubescrape
• Playlist extraction → use yt-dlp, pytube, or tubescrape

Features

• Video metadata: title, description, views, channel info, publish/upload dates, duration, tags, thumbnail
• Transcripts / captions: via timedtext URLs or the innertube get_panel endpoint, with automatic fallback
• Comments: via the innertube next continuation API, with deduplication and pagination
• Dislike counts: from the Return YouTube Dislike API
• Extractive summaries: word-frequency-based summarization of transcript or description text
• Access-block detection: detects CAPTCHAs, consent walls, and sign-in challenges (does not bypass them)
• Structured data models: typed dataclasses with JSON serialization
• CLI: convenient command-line interface
• Configurable: timeout, retries, delays, comment limits, language preference

Installation

Option 1: pip (requires Chrome installed locally)

pip install yt-network-scraper

Prerequisites: Google Chrome must be installed. Selenium Manager will automatically fetch a matching ChromeDriver in recent Selenium versions.

Option 2: Docker (no Chrome installation needed)

No need to install Chrome, Python, or any dependencies — Docker handles everything:

# Build the image
docker build -t yt-network-scraper .

# Scrape a video and save output to ./output/result.json
docker run --rm -v "$(pwd)/output:/output" yt-network-scraper \
  video "https://youtu.be/ALyQ-c9_HBI" --comments 25 --pretty --out /output/result.json

Option 3: Docker Compose

# Build and run with docker compose
docker compose run --rm yt-network-scraper \
  video "https://youtu.be/ALyQ-c9_HBI" --comments 25 --pretty --out /output/result.json

The docker-compose.yml is included in the repo. Output files are saved to the ./output/ directory via a mounted volume.

Quick Start

from yt_network_scraper import YouTubeScraper, ScraperConfig

config = ScraperConfig(max_comments=50, transcript_language="en")

with YouTubeScraper(config) as scraper:
    result = scraper.get_video("dQw4w9WgXcQ")

    print(result.metadata.title)
    print(f"Views: {result.metadata.views}")
    print(f"Channel: {result.metadata.channel_name}")

    if result.transcript.available:
        print(f"Transcript: {result.transcript.text[:200]}")

    if result.summary.available:
        print(f"Summary: {result.summary.text}")

    for comment in result.comments:
        print(f"  {comment.author}: {comment.text}")

You can pass a full URL, a youtu.be link, a shorts URL, or a bare 11-character video ID:

scraper.get_video("https://www.youtube.com/watch?v=dQw4w9WgXcQ")
scraper.get_video("https://youtu.be/dQw4w9WgXcQ")
scraper.get_video("https://www.youtube.com/shorts/dQw4w9WgXcQ")
scraper.get_video("dQw4w9WgXcQ")

Sample Response

Here is an example of the actual JSON output you get when scraping a real video. This was produced by running:

yt-network-scraper video "https://youtu.be/ALyQ-c9_HBI" --comments 5 --pretty

The result is a VideoResult object. Calling result.to_dict() (or using --pretty / --out in the CLI) produces JSON with this structure:

{
  "video_id": "ALyQ-c9_HBI",
  "source_url": "https://www.youtube.com/watch?v=ALyQ-c9_HBI&hl=en&persist_hl=1",
  "metadata": {
    "video_url": "https://www.youtube.com/watch?v=ALyQ-c9_HBI",
    "title": "এটিএন বাংলার সন্ধ্যা ৭ টার সংবাদ । 16.08.2026 | Today News ...",
    "description": "#atn #atnbangla #atnbanglanews ... Fair Usage Policy: ...",
    "views": 64649,
    "channel_name": "ATN Bangla News",
    "channel_id": "UCbgcYEdMsuypG2NJ-znBp3w",
    "channel_url": "http://www.youtube.com/@ATNBanglanews",
    "channel_subscribers": "10.3M subscribers",
    "upload_date": "2026-08-16T07:45:34-07:00",
    "publish_date": "2026-08-16T07:45:34-07:00",
    "duration_seconds": 2028,
    "category": "News & Politics",
    "is_live": false,
    "keywords": ["atn bangla news", "atnbangla", "bangla news", ...],
    "thumbnail": "https://i.ytimg.com/vi/ALyQ-c9_HBI/maxresdefault.jpg"
  },
  "engagement": {
    "comment_count_scraped": 5,
    "likes": 652,
    "views": 64649,
    "dislikes": {
      "source": "returnyoutubedislikeapi.com",
      "dislikes": 10,
      "likes": 648,
      "rating": 4.94,
      "view_count": 63292
    },
    "comment_count": 8
  },
  "transcript": {
    "available": true,
    "segments": [
      {
        "text": "আসসালামু আলাইকুম। এটিএন বাংলা সংবাদে সবাইকে স্বাগত ...",
        "start_ms": 8000,
        "duration_ms": null,
        "time": "0:08"
      },
      {
        "text": "রাজপথে বিশৃঙ্খলা সৃষ্টিকারীদের রুখে দেওয়ার আহ্বান ...",
        "start_ms": 16000,
        "time": "0:16"
      }
    ],
    "text": "আসসালামু আলাইকুম। এটিএন বাংলা সংবাদে সবাইকে স্বাগত ...",
    "language": "bn",
    "name": "Bangla (auto-generated)",
    "is_auto_generated": true,
    "source": "browser_network_get_panel"
  },
  "summary": {
    "available": true,
    "text": "আসসালামু আলাইকুম। এটিএন বাংলা সংবাদে সবাইকে স্বাগত ...",
    "method": "lead_sentences"
  },
  "comments": [
    {
      "comment_id": "Ugxxd7ztzYUkFIlchVl4AaABAg",
      "likes": 5,
      "reply_count": 0,
      "is_pinned": false,
      "is_hearted": true,
      "author": "@MdHoksap",
      "author_channel_id": "UC5DBcRWdv8oCmPhIauoy3gw",
      "author_channel_url": "/@MdHoksap",
      "text": "তারেক রহমান চাঁন্দাবাজের জন্য অপযোগী ...",
      "published": "3 hours ago"
    },
    {
      "comment_id": "UgzgNUoDsKC25nuDmhp4AaABAg",
      "likes": 0,
      "reply_count": 0,
      "is_pinned": false,
      "is_hearted": true,
      "author": "@MinhajUddin-s8n",
      "text": "SALARY OF ALL GOVT WORKERS IN SENIOR RANK MUST BE REDUCED ...",
      "published": "1 hour ago"
    }
  ],
  "network": {
    "access_status": {
      "blocked": false,
      "reasons": [],
      "message": "Access looks normal"
    },
    "api_key_found": true,
    "captured_event_count": 2256,
    "dom_scraping": false,
    "bot_evasion": false
  }
}

What each field contains

API Usage

YouTubeScraper

The main scraper class. Must be used as a context manager to manage the browser lifecycle.

from yt_network_scraper import YouTubeScraper, ScraperConfig

config = ScraperConfig(
    headless=True,           # Run Chrome in headless mode
    timeout=25,              # Browser page-load timeout (seconds)
    max_comments=25,         # Maximum comments to fetch
    transcript_language="en", # Preferred transcript language
    request_delay=1.5,       # Delay between fallback requests (seconds)
    max_page_retries=2,      # Retries on access-block pages
)

with YouTubeScraper(config) as scraper:
    result = scraper.get_video("VIDEO_ID")

VideoResult

The return type of get_video(). Contains:

Call result.to_dict() to serialize the entire result to a JSON-compatible dictionary.

Exceptions

from yt_network_scraper import (
    ScraperError,              # Base exception
    InvalidVideoURLError,      # URL/ID could not be parsed
    AccessBlockedException,    # YouTube returned an access challenge
    SeleniumNotInstalledError, # Selenium is not installed
    BrowserNotInitializedError, # Not used as a context manager
)

CLI Usage

# Scrape a video and print JSON to stdout
yt-network-scraper video "https://www.youtube.com/watch?v=VIDEO_ID"

# Save to a file with pretty-printing
yt-network-scraper video VIDEO_ID --out result.json --pretty

# Fetch up to 100 comments in French
yt-network-scraper video VIDEO_ID --comments 100 --lang fr

# Show Chrome (for debugging)
yt-network-scraper video VIDEO_ID --no-headless

# Custom timeout and retries
yt-network-scraper video VIDEO_ID --timeout 60 --retries 5

Configuration

All configuration is done through the ScraperConfig dataclass:

Package Architecture

The package is organized into focused, single-responsibility modules under src/yt_network_scraper/. This separation of concerns makes the codebase easy to test, maintain, and extend:

How a scrape works

Design principles

• Network layer is isolated — all Selenium and HTTP calls live in client.py. Parsing functions in parsing.py are pure and take dicts as input, making them trivial to test with fixtures.
• Typed models everywhere — the scraper never returns raw dicts. Every result is a typed dataclass with documented fields, optional fields are None when unavailable, and to_dict() produces clean JSON.
• Defensive parsing — YouTube changes payload shapes frequently. Every parser uses safe key lookups (find_key, find_all_keys) and returns None or empty lists for missing fields instead of raising exceptions.
• No bot evasion — the scraper detects access blocks but never tries to bypass them. If YouTube returns a CAPTCHA or consent wall, the network.access_status field reports it and the scrape completes with available data.
• Configurable behavior — ScraperConfig controls headless mode, timeout, retries, delays, comment limits, transcript language, and user agent. Sensible defaults work for most cases.

Error Handling

The scraper uses a typed exception hierarchy. All exceptions inherit from ScraperError:

from yt_network_scraper import YouTubeScraper, ScraperError

try:
    with YouTubeScraper() as scraper:
        result = scraper.get_video("bad_url")
except ScraperError as e:
    print(f"Scraping failed: {e}")

The scraper is designed to be defensive — YouTube frequently changes internal payload shapes. Missing fields are returned as None or empty lists rather than raising exceptions. The network.access_status field in the result indicates whether YouTube returned an access challenge.

Development

# Clone the repository
git clone https://github.com/DIP-RO/yt-network-scraper.git
cd yt-network-scraper

# Create a virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install in development mode with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run tests with coverage
pytest --cov=yt_network_scraper

# Build the package
python -m build

# Validate the build
twine check dist/*

Running Tests

pytest

The test suite uses mocked HTTP responses and mocked Selenium drivers — no live YouTube requests or browser instances are required. Tests cover:

• Package imports and public API surface
• Data model serialization
• URL parsing and ID extraction
• YouTube payload parsing (metadata, transcripts, comments)
• Network fetching with mocked sessions (success, errors, edge cases)
• Client orchestration with mocked Selenium
• CLI argument parsing and output

License

This project is licensed under the MIT License. See LICENSE for details.

Attribution

Developed and maintained by Dipro Paul.

Disclaimer

This package is intended for legitimate research, data analysis, and automation of publicly available YouTube video data. Users are responsible for complying with:

• YouTube's Terms of Service
• Applicable local and international laws
• Rate limiting and access restrictions

This package does not bypass CAPTCHAs, evade bot detection, circumvent authentication, or harvest credentials. If YouTube returns an access challenge, the scraper reports it rather than attempting to work around it. Use responsibly and at your own risk.