# Youtube > When a workflow needs public YouTube data through UnifAPI — channel, video, and playlist metadata, subscriber/view/like counts, video comments, caption tracks, channel community posts, related and trending videos, hashtag videos, and search across videos, channels, and playlists. Also use on "who owns this topic on YouTube," "pull this channel's videos," "YouTube view counts," "what are commenters saying," "get this video's captions/transcript," "this channel's playlists," or when another skill (creator vetting, competitor reception, content demand) needs the deterministic YouTube read path. Connect via the `unifapi` skill first. Read-only research, never uploads. - Skill: `unifapi-agent/youtube` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add unifapi-agent/youtube` - Raw SKILL.md: https://api.skillmd.com/api/skills/unifapi-agent/youtube/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - License: MIT - Author: unifapi-agent (https://skillmd.com/u/unifapi-agent) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/unifapi-agent/youtube --- # youtube The deterministic read path for public **YouTube** data through UnifAPI. This is a **Data Skill**: it does not run a marketing job on its own — it names the concrete `youtube/...` operations, response shapes, and the real coverage limits so any channel- or content-first workflow (creator vetting, competitor reception, content demand) reads from one known recipe instead of rediscovering the surface each time. Read-only — **eyes, not hands**. It researches public YouTube data and returns cited records; it never uploads, comments, or subscribes, and UnifAPI never holds YouTube credentials. ## Use the `unifapi` skill for live evidence Connect once through the shared **`unifapi`** skill (OAuth MCP), then call the operations below. Channels are keyed by **`{channel_id}`** and videos by **`{video_id}`** — resolve a handle or URL to a channel id first (see Gotchas). Keep any `billing` metadata so the output can state record cost. ## Response contract Single-entity endpoints return the object in `data`: ```json { "request_id": "unif_...", "data": {}, "billing": { "records_charged": 1, "balance_remaining": 99 } } ``` List endpoints return an array in `data` plus `pagination`: ```json { "request_id": "unif_...", "data": [], "pagination": { "has_more": false, "next_cursor": null }, "billing": { "records_charged": 1 } } ``` When `pagination.has_more` is true, pass `pagination.next_cursor` as the next request's `cursor`. Always preserve `billing` when reporting cost. ## Core operations | Need | Operation | | --------------------------- | -------------------------------------------------------------------------------------------- | | Search videos | `youtube/search` (`?q=...`, optional `sort_by`/`upload_date`/`duration`/`region`/`language`) | | Search channels / playlists | `youtube/search/channels` · `youtube/search/playlists` (`?q=...`) | | Trending | `youtube/trending` (optional `type` = now/music/gaming/movies) | | Hashtag videos | `youtube/hashtags/{tag}/videos` | | Resolve handle/URL → id | `youtube/resolve/channel-id` (`?url=...`) | | Channel page | `youtube/channels/{channel_id}` | | Channel videos / shorts | `youtube/channels/{channel_id}/videos` · `.../shorts` | | Channel playlists | `youtube/channels/{channel_id}/playlists` | | Channel community posts | `youtube/channels/{channel_id}/community` | | Search within a channel | `youtube/channels/{channel_id}/search` (`?q=...`) | | Video metadata | `youtube/videos/{video_id}` | | Related videos | `youtube/videos/{video_id}/related` | | Video comments | `youtube/videos/{video_id}/comments` (optional `sort_by` = top/newest) | | Video captions / transcript | `youtube/videos/{video_id}/captions` | | Playlist page / videos | `youtube/playlists/{playlist_id}` · `.../videos` | Need a field not listed here? Use the `unifapi` skill's `get_operation` to read the exact schema before calling — but pick the operation from this table, don't discover blind. ## Workflow The deterministic recipes. Pick the one that matches the job; each names exactly what to call. 1. **Frame the query.** Call `youtube/search?q=...` for who owns a topic, or `youtube/trending` for what's surging now. 2. **Resolve a channel.** Call `youtube/resolve/channel-id?url=...` to get the `channel_id`, then `youtube/channels/{channel_id}` for `subscriber_count`, `video_count`, and `view_count`. 3. **Read a channel's output.** Call `youtube/channels/{channel_id}/videos` (and `.../shorts`); each video's `view_count` is the demand signal. Page via `next_cursor`. `youtube/channels/{channel_id}/playlists` maps how the catalog is organized; `youtube/channels/{channel_id}/community` reads announcements and engagement outside videos; `youtube/channels/{channel_id}/search?q=...` finds a creator's coverage of one topic without scanning the whole catalog. 4. **Read a video and its neighborhood.** Call `youtube/videos/{video_id}` for metadata (now including `like_count`, `category`, `published_at`, `has_captions`), then `youtube/videos/{video_id}/related` to map adjacent demand and competing content. 5. **Read audience and transcript.** Call `youtube/videos/{video_id}/comments` (`sort_by` top/newest) for what the audience says — raw comment text, like and reply counts; any sentiment or FAQ grouping is yours to compute, the API returns text only. Call `youtube/videos/{video_id}/captions` for caption tracks (timed-text url per language) to read or analyze the transcript. 6. **Walk a playlist or hashtag.** Call `youtube/playlists/{playlist_id}` then `.../videos` to ingest a curated series in order; `youtube/hashtags/{tag}/videos` tracks a campaign or trend across creators. 7. **Size topic demand.** Aggregate `view_count` across the search and related results for a topic; `title`, `description`, and `keywords` show how the topic is framed. 8. **Cite everything.** Every figure ties back to the video or channel it came from; report `billing.records_charged` (or estimate when billing metadata is absent). ## Shape notes - **`YouTubeChannel`** — keyed by `{channel_id}`. `subscriber_count`, `video_count`, `view_count`, `handle`, `keywords`, `links`, `country`, `created_at`. `is_verified` is **not populated by the upstream and is always `false`** — don't report verification you can't read. - **`YouTubeVideo`** — keyed by `{video_id}`. `view_count`, `like_count` (best-effort; `0` when the creator hides it), `category`, `channel_id`, `author`, `duration_seconds`, `keywords`, `published_at`, `has_captions`, `is_live`, `is_private`. - **`YouTubeVideoPreview`** — search/related/list rows: adds `published_time`. `category` is part of the shape but empty on list rows (only `YouTubeVideo` detail carries it). - **`YouTubeComment`** — `text`, `author`, `author_channel_id`, `like_count`, `reply_count`, `published_time`, `is_verified`, `is_creator`. Text only — no sentiment field. - **`YouTubeCaptions`** — `video_id`, `tracks[]` (`language_code`, timed-text `url`, `is_translatable`), `translation_languages[]`. Single object, billed as one record. Track urls are time-limited. - **`YouTubePlaylist`** / **`YouTubePlaylistPreview`** — playlist `title`, `channel_id`, `author`, `video_count` (detail adds `view_count`, `last_updated`; preview adds `first_video_id`). - **`YouTubeChannelPreview`** — search-channels rows: `title`, `handle`, `description`, `subscriber_count`, `video_count`. - **`YouTubeCommunityPost`** — `text`, `author`, `vote_count`, `reply_count`, `attachment_type`, `image_url`. ## Gotchas - **Resolve only goes to channel ids.** `youtube/resolve/channel-id?url=...` resolves a handle or URL to a `{channel_id}` — call it **before** any `youtube/channels/{channel_id}/...` call. There is no video- or playlist-id resolver; pass a known `{video_id}` / `{playlist_id}` (parse it from the URL yourself). - **Comments are raw text.** `youtube/videos/{video_id}/comments` returns comment text plus like/reply counts — no sentiment, no FAQ grouping. Compute those yourself from the text; don't claim a sentiment field the API doesn't return. - **`like_count` is best-effort.** `YouTubeVideo.like_count` is `0` when the creator hides likes — treat `0` as "hidden or none," not authoritative. - **`is_verified` is always `false`** for channels (upstream doesn't expose it) — don't report channel verification. - **Captions can be empty.** A video with captions disabled returns `tracks: []`; check before promising a transcript. Track urls are time-limited. - **Hashtags** take the tag with or without a leading `#` (`youtube/hashtags/lofi/videos`). - A low balance can silently truncate list pages: check `billing.truncated_due_to_balance` — when true the page is partial. ## Output Return the records the calling workflow needs, each cited to its video or channel, plus a one-line cost note (`records_charged`). When this skill is used directly, a compact demand brief is the default: ```markdown **{Channel}** — {subscribers} subscribers, {videos} videos, {totalViews} views. Top videos for {topic}: {title} ({views} views). Related/competitors: {channels}. Evidence: {URLs}. Records: ~{N}. ``` ## Related skills - **kol-pricing**, **creator-shortlist**, **audience-fit-check** (Influencer Marketing) — price and vet YouTube channels on top of this read path. - **competitor-profiling** (Competitive Intelligence), **content-opportunity-brief** (Content Strategy) — reception and content-demand work that reads YouTube first. - **unifapi** — the shared data skill: connect MCP and look up exact schemas with `get_operation`.