Jonathon's AI Wiki

yt-dlp (CLI Audio/Video Downloader — Skill Dependency)

4 min read

Source: Yt Dlp Yt Dlp 2026 04 27

yt-dlp (github.com/yt-dlp/yt-dlp) is the de-facto command-line audio/video downloader for the open-source ecosystem — a feature-rich CLI with support for thousands of sites, forked from the inactive youtube-dlc, which itself branched from youtube-dl. It is the underlying YouTube/multi-platform extraction utility that several Claude skills (notably last30days-skill) call out to when they need transcripts, metadata, audio, or video from YouTube and adjacent platforms. Public domain (Unlicense) for the project itself; bundled binaries may include components under GPL and other licenses. Python 3.10+ (or PyPy 3.11+); cross-platform binaries for Windows, Linux, and macOS.

Key Takeaways

  • Not a Claude skill. yt-dlp is a stand-alone CLI. It enters this wiki as a dependency — Claude skills like last30days-skill (research aggregator) and claude-video (the /watch video-input skill) require it for video/transcript extraction, and several social/research workflows lean on it for audio/video pulls.
  • Thousands of sites. YouTube is the headline case, but the supported-sites list is in the thousands — Twitch, Vimeo, SoundCloud, Bandcamp, news sites, podcast hosts, and more. Most workflows that say “download a video from anywhere” reduce to a yt-dlp call.
  • Format selection + post-processing. Pick exact formats (-f bestvideo+bestaudio/best), embed metadata, extract subtitles, split chapters, and integrate with ffmpeg for re-encode/extract-audio steps. SponsorBlock chapter handling is built in.
  • Three release channels. Stable (default), nightly (latest fixes for sites that change frequently), and master (pre-release). yt-dlp -U updates in place; auto-update flags exist on most binary builds.
  • Cross-platform binaries. Windows x64 / x86 / ARM64; Linux x86_64 / aarch64 / armv7l (glibc + musl variants); macOS universal binary (10.15+). Unix-style installs on systems with Python 3.10+ or PyPy 3.11+.
  • Licensing nuance. The yt-dlp project itself is Unlicense (public domain). Pre-built executables bundle dependencies under MIT, Apache 2.0, BSD, and GPL — relevant for redistributing a binary in a closed-source product.
  • Automation-friendly. Standard stdin/stdout/exit-code interface, no daemon, no auth flow for public content. Drops cleanly into shell pipelines, cron jobs, Claude Code subagents, and skill-internal tool calls.

Implementation

Tool/Service: yt-dlp (Unlicense / public domain; bundled deps under various licenses)

Setup (pip — most portable):

python -m pip install -U yt-dlp

Setup (Homebrew, macOS):

brew install yt-dlp

Setup (binary, no Python required): Download the appropriate executable for your platform from yt-dlp releases. The macOS / Windows / Linux glibc builds include auto-update; the “unpackaged” builds do not.

Cost: Free. No API key, no rate-limited service tier.

Integration notes:

  • Pair with ffmpeg for any re-encoding, audio extraction, or chapter splitting — yt-dlp shells out to ffmpeg automatically when it’s on $PATH.
  • For Claude Code workflows that pull video transcripts, the call shape is typically yt-dlp --write-auto-sub --skip-download --sub-format vtt <url> — pulls the auto-generated transcript without downloading the video itself.
  • last30days-skill bundles yt-dlp as an optional dependency for YouTube coverage. Without it, the skill skips YouTube and notes “0 results” for that source.
  • claude-video’s /watch command shells out to yt-dlp for the download leg of every video URL it ingests. Required dependency, not optional — setup.py preflights it. Same yt-dlp install path serves both skills.
  • Update channel matters for site stability: when a target site changes its API or DRM, the fix usually lands in nightly within hours; stable can lag by days. For long-running automation, pin to nightly + monitor.

Try It

  • Install yt-dlp before installing last30days-skill — verify YouTube transcripts are returning content rather than the “skipped: yt-dlp not installed” path.
  • Use yt-dlp inside a Claude Code subagent to extract transcripts from a list of competitor YouTube videos, then synthesize themes — useful for AI Marketing and AI Video Content research workflows.
  • For video production research: pull source clips with yt-dlp -f bestvideo+bestaudio and process locally rather than scraping web pages.
  • Add yt-dlp -U to a scheduled task (e.g., daily) on any machine running Claude skills that depend on it — frequent site changes make staying current load-bearing.

Graph View

Backlinks