ytscrape — YouTube Scraper for Python (No API Key)

The fast, free, open-source Python YouTube scraper. Scrape YouTube search results, video metadata, channel info, comments, replies and transcripts — no API key, no quota, no Selenium, no browser.

Installation • Quick start • MCP server • Features • Use cases • Comparison • FAQ • Docs

⭐ If ytscrape saves you time, please star the repo — it helps other developers searching for a YouTube scraper find it.

from ytscrape import YouTube

with YouTube() as yt:
    for video in yt.search("python", max_results=5):
        print(video.title, video.url)

ytscrape talks to YouTube's internal InnerTube API — the same endpoints the web app uses — and turns responses into typed, frozen dataclasses with transparent pagination. Pure HTTP: no Selenium, no Playwright, no API key. Sync (YouTube) and async (AsyncYouTube) share the same surface.

⚠️ This library uses YouTube's private endpoints. Use it responsibly and at your own risk — the endpoints may change over time.

CLI demo

The CLI prints colourful boxed tables under a mini ▶ ytscrape wordmark, with a spinner while requests are in flight:

ytscrape search "python tutorial" --filter videos --max 10
ytscrape video https://www.youtube.com/watch?v=dQw4w9WgXcQ
ytscrape channel @RickAstleyYT
ytscrape comments https://youtu.be/dQw4w9WgXcQ --replies --sort newest --max 20

python -m ytscrape … works too. See the full CLI guide for more.

MCP server

An optional Model Context Protocol server lets AI clients (Claude Code, Claude Desktop, Cursor, VS Code, …) scrape YouTube through ytscrape — no API key, no quota:

pip install "ytscrape[mcp]"

{
  "mcpServers": {
    "ytscrape": {
      "command": "ytscrape-mcp"
    }
  }
}

Behind a proxy or an IP block, pass the proxy in args (or set HTTPS_PROXY in env instead):

{
  "mcpServers": {
    "ytscrape": {
      "command": "ytscrape-mcp",
      "args": ["--proxy", "http://user:pass@proxy.example.com:8080"]
    }
  }
}

All tools are read-only. Hard limits (not flags): max_results is a page size capped at 50 (a larger value is clamped, not fetched), and every payload is capped at 16 000 characters (truncated plus a hint). Continue with next_cursor — do not raise the limit to "get the rest". get_video / get_channel default to detail="concise". The server speaks stdio by default; use ytscrape-mcp --transport streamable-http for hosted setups. Do not expose HTTP mode to the internet without authentication. See the MCP guide for the limit table, proxy setup and client configs.

What can you scrape from YouTube?

Export everything to JSON or CSV in one line.

Why ytscrape?

• 🔑 No API key, no quota — nothing to register, no billing project.
• 🧊 No browser — pure HTTP only.
• 🧩 Typed models (Video, Channel, Comment, Transcript, …) + py.typed.
• 💬 Every comment — replies included; CommentSort.NEWEST does not hide any.
• 📄 Transparent pagination — just iterate; continuation tokens are handled for you.
• 🌍 Localisation — language (hl) and region (gl), validated ISO codes.
• ⚡ Sync & async — optional httpx extra for AsyncYouTube.
• 🖥️ CLI included — ytscrape search "python" --max 10.
• 📤 JSON / CSV export — video.to_json(), dumps_csv(results), or ytscrape search "python" --format json.

Use cases

• 📊 Data science & research — build YouTube datasets for analysis.
• 🤖 AI / LLM / RAG pipelines — pull YouTube transcripts as training or retrieval data.
• 💬 Sentiment analysis — scrape YouTube comments at scale.
• 📈 SEO & marketing — keyword research, competitor channel monitoring.
• 🧪 Academic studies — reproducible collection of video & channel metadata.
• 🛠️ Bots & automations — lightweight alternative to the YouTube Data API v3.

Installation

pip install ytscrape            # or: uv add ytscrape
pip install "ytscrape[async]"   # optional async API (httpx)
pip install "ytscrape[mcp]"     # optional MCP server (ytscrape-mcp)

Requires Python 3.10+. Runtime deps: requests, pycountry, defusedxml (+ httpx with the async extra, mcp with the MCP extra).

Quick start

from ytscrape import YouTube, SearchFilter, CommentSort

with YouTube(language="en", region="US") as yt:
    for video in yt.search("python", filter=SearchFilter.VIDEOS, max_results=20):
        print(video.title, "-", video.url)

    details = yt.video("https://www.youtube.com/watch?v=dQw4w9WgXcQ")
    print(details.title, details.channel, details.views, details.length_seconds)

    for comment in yt.comments(
        "https://youtu.be/dQw4w9WgXcQ",
        include_replies=True,
        sort=CommentSort.NEWEST,
        max_results=100,
    ):
        marker = "  ↳" if comment.is_reply else "-"
        print(f"{marker} {comment.author}: {comment.text}")

Async (same methods, await / async for):

import asyncio
from ytscrape import AsyncYouTube, SearchFilter


async def main() -> None:
    async with AsyncYouTube(max_concurrency=8) as yt:
        async for video in await yt.search(
            "python", filter=SearchFilter.VIDEOS, max_results=5
        ):
            print(video.title)


asyncio.run(main())

📂 Runnable sync + async snippets for every feature: examples/ · full guides: vsmutok.github.io/ytscrape

Feature coverage

ytscrape vs. the alternatives

See the Migration Guide from scrapetube for a detailed comparison and code examples.

How it works

High-level flow — sync and async share the same models and parsing layer:

graph LR
    User[User code / CLI] --> Facade[YouTube / AsyncYouTube]
    Facade --> Client[InnerTubeClient / AsyncInnerTubeClient]
    Facade --> Results[Lazy results / comment threads]
    Client --> InnerTube[YouTube InnerTube API]
    Client --> Context[InnerTube context]
    Results --> Client
    Client --> Parsing[parsing helpers]
    Results --> Parsing
    Parsing --> Models[Frozen dataclasses]
    Models --> User

1. Load youtube.com once and extract the InnerTube context (API key, client version, visitor data).
2. POST to youtubei/v1/search, player, browse, next, etc. with that context.
3. Parse responses into frozen dataclasses; continuation tokens are followed while you iterate.

Documentation

Deep dives live on the docs site (not duplicated here):

FAQ

proxy = {"https": "http://user:pass@proxy:8080"}
details = YouTube(proxies=proxy).video("dQw4w9WgXcQ")

Contributing

Bug reports, ideas and PRs are welcome — see CONTRIBUTING.md.

git clone https://github.com/vsmutok/ytscrape && cd ytscrape
uv sync --dev
uv run pytest
uv run pre-commit run --all-files

• 🗺️ todo: todo.md
• 📝 Changelog: CHANGELOG.md

Star History

License

MIT — free for personal and commercial use.

_{Keywords: youtube scraper, python youtube scraper, scrape youtube, youtube scraping, youtube comments scraper, youtube transcript downloader, youtube search api python, youtube data api alternative, youtube api without key, innertube api, youtube metadata extractor, youtube channel scraper, youtube crawler, youtube shorts scraper, async youtube scraper, scrapetube alternative.}