yt-search-python v2.2.1

Search YouTube videos, playlists, channels, comments, transcripts, recommendations, suggestions, and stream metadata without the YouTube Data API v3.

• Python 3.9+
• Sync API: youtubesearchpython
• Async API: youtubesearchpython.future
• httpx>=0.28.1
• No YouTube Data API key or quota

Installation

pip install yt-search-python

Search

from youtubesearchpython import VideosSearch

search = VideosSearch("Arijit Singh", limit=10)
print(search.result())
search.next()
print(search.result())

Live-only search is optional and backward compatible:

live = VideosSearch("news", limit=10, is_live=True)
print(live.result())

Async search loads its first page on the first await next() call:

import asyncio
from youtubesearchpython.future import VideosSearch

async def main():
    search = VideosSearch("Arijit Singh", limit=10)
    first = await search.next()
    print(first)
    second = await search.next()
    print(second)

asyncio.run(main())

Video

from youtubesearchpython import Video

info = Video.getInfo("pnxL4OOzPEc")
formats = Video.getFormats("pnxL4OOzPEc")

PO token and visitor data can be supplied when required by the selected YouTube client/session:

formats = Video.getFormats(
    "pnxL4OOzPEc",
    po_token="YOUR_PO_TOKEN",
    visitor_data="YOUR_VISITOR_DATA",
)

ResultMode.dict and ResultMode.json are supported by video APIs.

StreamURLFetcher

StreamURLFetcher no longer uses yt-dlp. It can process a Video.getFormats() result or fetch the format data from a video ID/link itself.

from youtubesearchpython import StreamURLFetcher

fetcher = StreamURLFetcher(po_token="YOUR_PO_TOKEN", visitor_data="YOUR_VISITOR_DATA")
url = fetcher.get("pnxL4OOzPEc", 18)
all_streams = fetcher.getAll("pnxL4OOzPEc")

Existing dictionary input remains supported:

from youtubesearchpython import Video, StreamURLFetcher

formats = Video.getFormats("pnxL4OOzPEc")
result = StreamURLFetcher().getAll(formats)
print(result["streams"])
print(result["unresolved"])

Direct URLs and cipher entries that already contain a usable signature are returned without yt-dlp. Formats that still require YouTube's encrypted player-JavaScript signature deciphering are returned under unresolved instead of being presented as working URLs. URLs that still contain an n parameter are marked with throttled=True so callers can make an informed choice rather than silently receiving a falsely-deciphered URL.

Playlists

Regular playlists and YouTube Mix/Radio playlists (RD...) use YouTube's native Innertube endpoints.

from youtubesearchpython import Playlist

normal = Playlist.get("PLRBp0Fe2GpgmsW46rJyudVFlY6IYjFBIK")
mix = Playlist.get("https://youtube.com/playlist?list=RDpnxL4OOzPEc&playnext=1")

Mix results preserve YouTube's returned song order. Duplicate video IDs are removed without re-sorting the result. Generic comment/engagement continuation tokens from /next responses are not treated as playlist continuations.

For regular playlists, instantiate Playlist(link) and call getNextVideos() for continuation pages.

Recommendations

from youtubesearchpython import Recommendations

related = Recommendations.get("pnxL4OOzPEc")

Recommendation results preserve YouTube's response order, skip the source video, remove duplicate video IDs stably, and normalize thumbnails against each video's ID.

Suggestions

from youtubesearchpython import Suggestions

print(Suggestions.get("Guru Randhawa"))

YTS_PROXY and YTS_IDENTITY_TOKEN environment variables are supported by the suggestions transport.

Comments, transcripts, channels and hashtags

from youtubesearchpython import Comments, Transcript, Channel, Hashtag

comments = Comments.get("pnxL4OOzPEc")
transcript = Transcript.get("pnxL4OOzPEc", params="en")
channel = Channel.get("UC_x5XG1OV2P6uZZ5FSM9Ttw")
hashtag = Hashtag.get("music", limit=10)

Transcript retrieval first uses YouTube's native caption/player flow. The optional transcript extra keeps the legacy yt-dlp caption fallback available:

pip install 'yt-search-python[transcript]'

HTTP lifecycle

The library uses one canonical httpx transport layer. Idle keep-alive retention is disabled to avoid stale pooled sockets in long-running bots/services.

Sync shutdown:

# Optional forced teardown only
from youtubesearchpython import close_clients
close_clients()

Async shutdown:

# Optional forced teardown only
from youtubesearchpython.future import aclose_clients
await aclose_clients()

Proxy requests use scoped clients that are closed after each request. Temporary downloaded cookie files are also ownership-tracked and cleaned without deleting user-owned cookie files.

Main API

Search:

• Search
• VideosSearch
• ChannelsSearch
• PlaylistsSearch
• CustomSearch
• ChannelSearch

Content:

• Video
• Playlist
• Channel
• Comments
• Transcript
• Hashtag
• Suggestions
• Recommendations

Streaming:

• StreamURLFetcher

Utilities:

• ResultMode
• SearchMode
• VideoUploadDateFilter
• VideoDurationFilter
• VideoSortOrder
• ChannelRequestType

Compatibility notes

• youtubesearchpython.future is the supported async namespace.
• Legacy SearchVideos and SearchPlaylists imports remain available.
• YouTube's internal response structures and anti-abuse requirements can change without notice.
• A PO token does not replace player JavaScript signature or n-challenge transformation when YouTube requires those for a format.

License

MIT. See LICENSE.

Current maintainer: Prakhar Shukla / BillaSpace. Original project by Hitesh Kumar Saini (alexmercerind).

Optional PO token environment variables

Video and StreamURLFetcher keep their existing arguments, but can also read credentials from the environment when explicit values are not passed:

export YT_PO_TOKEN="..."
export YT_VISITOR_DATA="..."

Aliases YOUTUBE_PO_TOKEN and YOUTUBE_VISITOR_DATA are also supported. Explicit function/class arguments always take precedence over environment values.

Python compatibility

• Python 3.9+
• Runtime-tested on Python 3.13.5
• Audited against Python 3.14 asyncio removals/deprecations; the library uses asyncio.get_running_loop() and does not depend on the deprecated event-loop policy APIs.
• HTTP transport uses the tested httpx>=0.28.1,<1.0 range.

HTTP clients are managed internally. Normal sync applications require no explicit shutdown call, and async clients are closed automatically when their owning event loop shuts down gracefully (including asyncio.run()). close_clients() and aclose_clients() remain available only for optional forced teardown, tests, or unusual lifecycle control.