I keep running into people who reach for cobalt when they want to add "download this video or audio URL" to an app, hit a wall, and then go looking for what to use instead. Here's the short version of what I've figured out, because the cobalt-vs-yt-dlp decision usually isn't really about which downloader is better - it's about where you're running it.

cobalt is genuinely nice. Clean web UI, paste a link, get a file, nothing to install. For grabbing one video for yourself it's lovely and I'd just use it. The friction shows up the moment you want to use it programmatically.

The public cobalt API isn't a public API

This is the part that catches people out. cobalt has an API, but its own docs are explicit that the hosted instance is not a backend you can build other projects on. From cobalt's API docs:

hosted api instances (such as api.cobalt.tools) use bot protection and are not intended to be used in other projects without explicit permission

and

if you want to use the cobalt api, you should host your own instance or ask an instance owner for access.

So you can't just point your app at api.cobalt.tools. The path cobalt points you to is running your own instance - which is fine for personal use, but it moves the whole job onto your own box, and that's where the next problem shows up.

Self-hosting moves the problem, it doesn't remove it

Whether you self-host cobalt or run yt-dlp directly, once it's on a cloud server the hard part isn't the downloader, it's the IP. YouTube and friends throw a lot more bot checks at datacenter IP ranges (AWS, GCP, Railway, Fly, Hetzner), so a setup that works on your laptop starts returning "Sign in to confirm you're not a bot" or HTTP 403 the moment it runs in the cloud. That's why there are so many "fix cobalt on YouTube" and "cobalt Railway template" threads - and it's the exact same failure yt-dlp users hit on a server.

The yt-dlp equivalents of what you use cobalt for

If you're switching because you want a scriptable CLI, here are the direct commands. All verified on current yt-dlp (2026.10.26):

# Download a video (best quality, like cobalt's default) yt-dlp "https://www.youtube.com/watch?v=VIDEO_ID" # Audio only as MP3 (cobalt's "audio" mode) - needs ffmpeg yt-dlp -x --audio-format mp3 "https://www.youtube.com/watch?v=VIDEO_ID" # Cap the quality (cobalt's quality picker) yt-dlp -S "res:1080,ext:mp4" "https://www.youtube.com/watch?v=VIDEO_ID" # A whole playlist at once (cobalt does one link at a time) yt-dlp -o "%(playlist_index)02d-%(title)s.%(ext)s" "PLAYLIST_URL" 

Batch and scheduled downloads are where a CLI pulls ahead of a one-link-at-a-time UI, and yt-dlp's extractor list runs to well over a thousand sites.

When cobalt is still the right call

Not a takedown - cobalt is a well-made project. If you want to grab one video without opening a terminal, its hosted UI is faster and friendlier than installing a CLI, and you can self-host a private instance. For occasional personal downloads it's a perfectly good answer. yt-dlp pulls ahead when coverage, control, and scripting matter.

If the reason was "download inside an app/agent"

If you were hosting cobalt or yt-dlp specifically to pull media from inside an app, a cron job or an agent, the IP blocking above is the part that keeps breaking, not the downloader. fwiw I ended up building importly.io for exactly this - a hosted API where you paste a URL and get the file, transcript and metadata back, running from servers and agent sandboxes where datacenter IPs get blocked. $3 of free credit, no card, if you want to try it.

Full version with the server-side fixes is in the original guide: https://ytdlp.org/guides/yt-dlp-vs-cobalt