Links from YouTube, Twitter/X, Instagram, TikTok, Twitch, Vimeo, Facebook, Bilibili, VK, Dailymotion and Reddit all work with the video tools; only search is YouTube-only. It does not download video or audio for you — what comes back is transcript text, metadata and single frames, which is also why a two-hour talk is answerable without waiting on a download.
A video-reading server: transcripts, subtitle files, chapters, metadata and individual frames from links on eleven platforms, available as a hosted endpoint or self-hosted from a Docker image.
- `get_transcript` for the spoken content, `get_raw_subtitles` for an SRT file, `get_available_subtitles` to see which languages exist
- `get_video_info` — publisher, view count, likes, description and the caption languages
- `get_video_chapters`, so "go to the part about pricing" resolves to a timecode
- `get_video_frame` — the screen at a timecode you name
- `get_playlist_transcripts` over a range of a playlist, and `search_videos` for finding videos on YouTube with date filters and paging
- Long transcripts returned in parts, each response carrying a cursor for the next, so no text is lost
- Interactive widgets for `get_transcript`, `get_video_info`, `get_video_frame` and `search_videos` in clients that support MCP Apps
Nothing to install for the hosted route: add `https://gateway.mcpal.io/mcp/transcriptor` to your client and approve the browser sign-in. To self-host, `docker run --rm -p 4200:4200 artsamsonov/transcriptor-mcp:latest` serves Streamable HTTP on port 4200 and needs no account; the tools are identical.
One command — docker run --rm -p 4200:4200 artsamsonov/transcriptor-mcp:latest