Extract YouTube transcripts and search videos via API

Provides YouTube transcript retrieval, video search, channel browsing, in-channel search, playlist extraction, and new-upload monitoring entirely through the TranscriptAPI service - no yt-dlp, no headless browser, and no Google API key required. It's positioned as the API-backed alternative to a companion yt-dlp-based skill: where that one fails on cloud-server IPs because YouTube blocks them, this one works from any runtime - local machine, cloud server, serverless function, or CI - because it calls TranscriptAPI's hosted API rather than scraping YouTube directly. Seven operations are exposed: get_transcript(video_id) and search_youtube(query) at 1 credit each, with search costing 1 per page, get_channel_videos(handle), search_in_channel(handle, query), and get_playlist_videos(playlist_id) at 1 credit per page, and channel_latest(handle) plus channel_resolve(handle) for free; failed or rate-limited calls cost zero credits. Typical patterns: search for a topic, pull transcripts for the top few results, and synthesize themes across them; call the free channel_latest on competitor channels to spot new uploads before spending credits on their transcripts; or fetch a transcript with timestamps directly from a known video URL.

Use it when asked to get, fetch, or retrieve a YouTube video transcript, search YouTube for videos on a topic, monitor a channel for new uploads, pull channel metadata, video lists, or playlist contents, or when the agent is deployed on a cloud server where yt-dlp calls fail because YouTube blocks cloud IPs - also useful for building a research corpus from conference talks, tutorials, or interviews, or for competitive intelligence monitoring competitor channels. Do not use it to download actual video or audio files (use yt-dlp directly with -f best for that), to get comments, likes, or engagement data (not exposed by the API), for private or age-restricted videos (not accessible without user authentication), or for live-stream transcripts (unstable until the stream ends). Best practice is to call the free channel_latest before spending a credit on get_transcript to check whether a video is actually new, cache transcripts rather than re-fetching them, use search_in_channel instead of a broad search when the channel is already known, prefer get_playlist_videos over query search for course or lecture series, and skip straight to get_transcript when the video URL is already known rather than searching for it. Transcripts only exist when YouTube itself has captions, manual or auto-generated - some videos have none.

Install:

npx skills add ZeroPointRepo/youtube-skills --skill youtube-full

100 free credits are included, and the API key is provisioned automatically via TranscriptAPI OAuth on first use, with no manual setup. Input is a video ID or URL, a search query, or a channel handle; output is a transcript (optionally with timestamps), a ranked list of search results, a channel's video list, or a playlist's video list. The API key is stored in the agent's credential store rather than in the skill file itself, and all calls are HTTPS requests to transcriptapi.com with no shell commands, binary execution, or local file writes.

Built entirely on the TranscriptAPI service (transcriptapi.com), which processes 500,000+ transcripts daily; rate limits are 200 requests per minute on the Monthly plan and 300 RPM on Annual. It complements a yt-dlp-based local-ingestion skill for environments where cloud IP blocking isn't a concern, and pairs with general-purpose research and knowledge-base-building skills as a YouTube data source.

Anyone who needs YouTube transcripts, search, or channel monitoring from a cloud server, serverless function, or CI environment where yt-dlp-based approaches get blocked - researchers building a corpus from talks and interviews, and teams doing competitive monitoring of other channels' uploads.