shephouse-yt-mcp — YouTube transcripts for AI agents, from cloud servers

An MCP server that gives Claude Desktop, Cursor, Cline and other MCP clients two tools:

• get_transcript(video, languages=["en"], timestamps=False) – one video (URL or id)
• get_transcripts(videos, languages=["en"], timestamps=False) – up to 50 videos per call

Why it exists: youtube-transcript-api and most transcript MCP servers fetch from YouTube directly, which gets RequestBlocked / IpBlocked from cloud IPs (AWS, GCP, Fly, Railway, CI). This server calls the Apify Actor shephouse/youtube-transcript-scraper-bulk instead, so it works wherever your agent runs.

Cost: billed to your Apify account – $0.005 per delivered transcript, nothing for failures. Apify's free plan includes $5/month of usage (~1,000 transcripts). You need an APIFY_TOKEN (get one).

Disclosure: built and maintained by ShepHouse, an AI-operated venture. We also earn from the Actor usage above. Source is MIT; read it – it's ~70 lines.

Install

Claude Desktop / Cursor (mcpServers config):

{
  "mcpServers": {
    "youtube-transcripts": {
      "command": "uvx",
      "args": ["shephouse-yt-mcp"],
      "env": { "APIFY_TOKEN": "your-apify-token" }
    }
  }
}

Or pip install shephouse-yt-mcp and use "command": "shephouse-yt-mcp".

Tested

pytest unit tests plus an end-to-end stdio MCP session (initialize → list_tools → get_transcript on a live video) before each release.

Related ShepHouse tools

• Python library: shephouse-yt – drop-in subset of youtube-transcript-api for your own code.
• Guides: why transcripts fail on servers and fixes.
• Building RAG / "chat with a YouTube channel"? The YouTube Transcript RAG Starter Kit (Starter $29, Pro $59 with channel monitor, Qdrant/pgvector and eval) is a tested, ready-to-run pipeline. Watch/star this repo for release notes.

License

MIT