A YouTube MCP server most commonly fetches video transcripts so an AI client can summarise, extract quotes, or analyse content without you copying anything. The transcript path needs no API key. Here is the setup.
Setup — transcript server (no API key)
A TypeScript transcript server runs via npx and works in Claude Desktop, Cursor, Cline and others:
{
"mcpServers": {
"youtube": {
"command": "npx",
"args": ["-y", "@sinco-lab/mcp-youtube-transcript"]
}
}
}
Package names differ between implementations — check the repo you choose and swap the args. Running locally means transcript requests come from your machine, not a shared proxy. Restart the client; see MCP config file location.
Optional — YouTube Data API for more
Transcripts, chapters and URL parsing typically work key-free. For search, comments, channel and playlist data, some servers want a YouTube Data API v3 key (free, from Google Cloud Console) passed via env.
Verify
Ask: "Get the transcript of this YouTube video and give me 5 key takeaways:
Common problems
- No transcript found — the video has captions disabled, or is region-locked.
- Quota errors — only for API-key features; raise quota in Google Cloud or stick to transcripts.
- No tools in client — Node 18+ for
npx; see Cursor MCP not working.
Going further
YouTube transcripts pair with a filesystem server (save notes) and a web-search server for research workflows. Browse the search category or curated loadouts.