Skip to content

use-case-n8n-automation.md

Latest commit

 

History

History
79 lines (57 loc) · 4.33 KB

use-case-n8n-automation.md

File metadata and controls

79 lines (57 loc) · 4.33 KB

Use case: No-code automation (n8n)

Use Transcriptor MCP from n8n (or other MCP clients) to build workflows that search videos, fetch transcripts, and process playlists — without writing code.

Who this is for

  • You build automation workflows in n8n or similar tools that support MCP (Model Context Protocol).
  • You need pipelines like: search videos → get transcript → summarize or analyze, or batch process a playlist and get all transcripts.
  • You want streamable HTTP so the MCP server runs on a host (e.g. your server or Docker) and n8n connects by URL.

Connect n8n to Transcriptor MCP

  1. Run the MCP server over HTTP (streamable HTTP on port 4200):

    • Docker: docker run -p 4200:4200 -e MCP_PORT=4200 -e MCP_HOST=0.0.0.0 artsamsonov/transcriptor-mcp npm run start:mcp:http
    • Or use docker-compose / your own host; see quick-start.mcp.md.

  2. In n8n, add the MCP server (MCP Client Tool or equivalent):

    • MCP Server URL: http://<host>:4200/mcp (streamable HTTP).
    • If your server uses MCP_AUTH_TOKEN, configure the client to send Authorization: Bearer <token>.

  3. Proxy note: If n8n runs behind a reverse proxy that sets X-Forwarded-For, you may need to set N8N_PROXY_HOPS (e.g. 1) so the server does not reject requests. See the main README for details.

Example workflows

Workflow 1: Search → transcript → summarize

  1. Trigger (schedule, webhook, or manual).
  2. MCP tool: search_videos
    • query: e.g. "product launch keynote 2024"
    • limit: 5
    • Output: list of videos with videoId, title, url, uploader, viewCount.
  3. Process first result (e.g. get url from first item).
  4. MCP tool: get_transcript
    • url: from step 2.
    • Output: cleaned plain text (first chunk).
  5. Send to LLM or another node for summarization, extraction, or storage.

For full transcript (long videos), use get_raw_subtitles with response_limit and repeat with next_cursor until is_truncated is false.

Workflow 2: Playlist transcripts in batch

  1. Trigger with a playlist URL (e.g. https://www.youtube.com/playlist?list=...).
  2. MCP tool: get_playlist_transcripts
    • url: playlist URL (or watch URL with list= parameter).
    • Optional: playlistItems: "1:10" (first 10), maxItems: 20, lang: "en", type: "auto".
    • Output: results array of { videoId, text }.
  3. Loop or process each item (e.g. save to DB, send to NLP, or aggregate for a report).

You can combine with get_video_info per videoId if you need metadata (title, channel, duration) alongside the transcript.

Workflow 3: Raw subtitles with pagination

For long videos where you need the full SRT/VTT or exact timestamps:

  1. MCP tool: get_raw_subtitles
    • url: video URL
    • response_limit: 50000 (default; max 200000)
    • Optional: type, lang, format (srt, vtt, ass, lrc).
  2. If the response has is_truncated: true, call get_raw_subtitles again with the same url and next_cursor from the previous response.
  3. Repeat until is_truncated is false, then concatenate or store the chunks.

Tools reference (n8n)

Tool Typical use in n8n
search_videos Find videos by query; optional uploadDateFilter, matchFilter, limit, offset.
get_transcript Plain-text transcript (URL only; auto language). Good for summaries and LLM input.
get_raw_subtitles Raw SRT/VTT with pagination (response_limit, next_cursor) for long content.
get_playlist_transcripts Batch transcripts for many videos from one playlist URL.
get_video_info Metadata (title, channel, duration, tags, etc.) for SEO or filtering.
get_available_subtitles List official/auto languages before choosing type/lang in get_raw_subtitles.

All tools accept a video URL from any supported platform (YouTube, Twitter/X, Instagram, TikTok, Twitch, Vimeo, Facebook, Bilibili, VK, Dailymotion, Reddit) or a YouTube video ID.

See also