Paste a video or post URL: download it, pull the caption, transcribe the audio locally or in the cloud, and research it. Reuses OmniGet's managed yt-dlp, ffmpeg, cookies and models when the app is installed.

GitHub stars
14.5K

Install

/plugin install omniget@omniget

Run it inside Claude Code after adding the marketplace with /plugin marketplace add tonhowtf/omniget.

Contents

Discovered from source

Commands 6

Connect this Claude Code to OmniGet's local MCP server.

  1. Get the current steps from the OmniGet app on this computer. It listens on one of the ports 47720 to 47729; use the first answer that starts with # Connect:

    for p in 47720 47721 47722 47723 47724 47725 47726 47727 47728 47729; do r=$(curl -s -m 2 "http://127.0.0.1:$p/connect?client=claude"); case "$r" in "# Connect"*) echo "$r"; break;; esac; done

    (Windows PowerShell: try curl.exe -s "http://127.0.0.1:<port>/connect?client=claude" on each port.) If no port answers that way, OmniGet is not open or is older than this feature: ask the person to open or update it (https://getomniget.com/download) and try again.

  2. Follow the steps it returns. If the user wants to download courses or sites behind a login, or to write a connector, add --data-urlencode "browser=1" to the request so the Allow also offers the Built-in Browser. The user clicks Allow in the OmniGet window; OmniGet writes the omniget server into Claude Code's config itself, so never ask for a token.

Run bash "${CLAUDE_PLUGIN_ROOT}/scripts/doctor.sh" --check-keys and explain the result in plain words: what works today (captions, local Whisper, Gemini, OpenAI), what is missing, and the exact command that would add each missing piece. Do not run any install command unless the user answers yes.

If tools or keys are missing, tell the user they can fix it in one step with /omniget:setup (or, for a key, by running keys.sh set in their own terminal).

Download the media for the arguments below using the omniget-fetch skill.

Run:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/fetch.sh" $ARGUMENTS

Then report the saved file path, size and duration in one or two sentences. If the script fails, run bash "${CLAUDE_PLUGIN_ROOT}/scripts/doctor.sh", apply the error table from the omniget-fetch skill, and ask before installing anything.

Build a research note for the URL in the arguments using the omniget-transcribe skill.

  1. bash "${CLAUDE_PLUGIN_ROOT}/scripts/research.sh" "<url>" for title, uploader, duration and caption.
  2. bash "${CLAUDE_PLUGIN_ROOT}/scripts/transcribe.sh" <arguments> for the transcript.
  3. Read the transcript file. Write <transcript path without .md>.notes.md containing: title and source link; the caption verbatim when present; a five-line summary; key points with [mm:ss] references; notable quotes with timestamps; open questions worth checking.
  4. Reply with the note's path and the summary. Mention which backend produced the transcript.

Set up everything the skill needs, using the omniget-fetch / omniget-transcribe skills' scripts.

  1. Run the setup script with any arguments the user passed:
    bash "${CLAUDE_PLUGIN_ROOT}/scripts/setup.sh" $ARGUMENTS
    
    • It checks for yt-dlp and ffmpeg, lists anything missing, installs it after one confirmation (Mac/Windows), and downloads omniget-cli (native Instagram/X/Bilibili extractors). --yes skips the prompt; --local adds a local Whisper engine + model; --no-cli skips omniget-cli; --update refreshes yt-dlp and omniget-cli.
    • It cannot type the user's API keys, and neither can you. It reports which keys work and prints the exact command for the user to run in their own terminal to add one.
  2. Read the output and tell the user, in plain words: what is now installed, which transcription options are available to them (captions always; local if a model is present; Gemini/OpenAI if a key tested "working"), and — only if they have no working key and no local model — that to add a key they should run, in their own terminal: bash "${CLAUDE_PLUGIN_ROOT}/scripts/keys.sh" set
  3. Do not ask the user to paste a key to you, and never run keys.sh set yourself — it needs their terminal and their keys must not pass through this conversation.

Transcribe the media in the arguments below using the omniget-transcribe skill.

  1. If --summarize is present, remove it from the arguments and remember to summarize.
  2. For a URL, first run bash "${CLAUDE_PLUGIN_ROOT}/scripts/research.sh" "<url>" to get the title and caption.
  3. Run bash "${CLAUDE_PLUGIN_ROOT}/scripts/transcribe.sh" <remaining arguments>.
  4. Read the transcript file named in the JSON output.
  5. Reply with the transcript path, the backend used, and, when asked, a summary with [mm:ss] references. Otherwise give a two-sentence description of the content.

Sponsors

Details

Marketplace
tonhowtf/omniget
Source
claude-plugin/omniget
Category
Productivity
Keywords
omniget, yt-dlp, video, transcript, whisper, download