If you want an AI assistant to work with YouTube — summarize a talk, pull quotes, or research a topic across many videos — you first have to get the transcript into the model. This is the hub guide: it covers every way to do that across the major assistants, links to step-by-step setups, and helps you pick the right path.

The two ways in: MCP or API

There are two integration styles, and which you use depends on the assistant:

MCP (Model Context Protocol). A small local program the assistant launches and calls. Best for desktop assistants — Claude Desktop, Cursor, Windsurf. One package, scribefy-mcp, gives the assistant four YouTube tools, and it's listed in the official MCP registry for a one-line install.
REST API. Plain HTTPS endpoints with a Bearer key. Best for ChatGPT Custom GPTs (via Actions) and for your own code. Same engine as the MCP server, just over HTTP.

Both return the same thing: a clean, timestamped transcript an AI can quote and cite.

The four tools

Whether over MCP or API, Scribefy exposes the same capabilities:

extract_transcript — a video's full timestamped transcript
search_videos — free-text YouTube search
get_video_metadata — title, channel, duration, caption tracks
get_related_videos — the "Up next" feed

Three are free; only extraction spends credits, and cached videos are free. That mix is what makes research affordable: the assistant searches and triages for free, then extracts only the videos worth reading.

Set it up for your assistant

Pick your tool — each has a step-by-step guide:

Claude Desktop → Build an AI workflow with Claude
Cursor → YouTube transcripts in Cursor
Windsurf → YouTube transcripts in Windsurf
ChatGPT → YouTube transcripts in ChatGPT
Your own code or agent → the developer guide

All three desktop assistants use the identical MCP config:

{
  "mcpServers": {
    "scribefy": {
      "command": "npx",
      "args": ["-y", "scribefy-mcp"],
      "env": { "SCRIBEFY_API_KEY": "sk_live_…" }
    }
  }
}

Only where the config lives differs (Claude's developer settings, Cursor's mcp.json, Windsurf's mcp_config.json). Grab the key from the dashboard.

What you can do once it's wired in

Summarize a video with timestamped key points. (Summarize a video with Claude.)
Pull quotes with exact timestamps for citations.
Research a topic across many videos — search, triage, extract, synthesize. (Analyze videos with Claude.)
Repurpose a talk into notes, an outline, or a blog draft.

The research pattern — search → filter → extract — is the real unlock: the assistant finds candidates (search_videos), checks them (get_video_metadata), and extracts only the keepers (extract_transcript). Free until the extract step, cached after.

Why timestamps matter for AI

Scribefy segments every transcript with timestamps, so an assistant can cite where a claim was made ("at 12:40 she argues…") instead of paraphrasing the whole video. That makes answers verifiable and source-linked — and you can jump straight to the moment.

Which option should you pick?

On Claude, Cursor, or Windsurf? Use the MCP server — least setup, and the assistant drives the tools itself.
On ChatGPT? Use a Custom GPT Action on the REST API.
Building your own app or agent? Call the REST API directly, or run the MCP server in your stack.
Deciding between Scribefy and the alternatives? See the honest comparison in choosing a transcript MCP for AI agents.

Frequently asked questions

Is it really the same across assistants?

Yes. scribefy-mcp (MCP) and the REST API share one engine, so transcripts, search, and metadata are identical no matter how you connect.

What does it cost?

New accounts get 2 free credits. Search, metadata, and related-videos are free; only extraction spends a credit (1–8 by length), and cached videos are free. See pricing.

What are the limits?

Scribefy is YouTube-only and needs a video to have captions — human or auto-generated. It doesn't run its own speech-to-text. For that case, see transcripts without captions.

Can I get transcripts in other languages?

Yes, when the video has captions in that language. See transcripts in any language.

Pick your assistant above, grab a key from the dashboard, and give your AI YouTube as a research source.