Cursor is great at reasoning over code — but it can research video content too, if you give it the right tool. With the Scribefy MCP server, Cursor's agent can search YouTube, inspect videos, and pull full timestamped transcripts on its own, without you leaving the editor. Here's the setup and what it unlocks.

What you get

The Model Context Protocol (MCP) lets Cursor call external tools. Scribefy's server (scribefy-mcp) adds four:

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 — YouTube's "Up next" feed

Three of the four are free; only extraction spends credits, and cached videos are free. For how this compares to other options, see choosing a transcript MCP for AI agents.

Setup

You'll need a Scribefy API key from the dashboard (the API + MCP plan). Then add the server to Cursor's MCP config — either way works:

Option A — edit the config file (stable across versions). Create or open ~/.cursor/mcp.json (global, all projects) or .cursor/mcp.json in a single project, and add:

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

Option B — use the UI. Open Cursor Settings → MCP (labelled Tools & Integrations in newer builds) → Add new MCP server. That opens the same mcp.json; paste the scribefy block above.

Save, then make sure the server is enabled — Cursor shows a green dot and lists the four tools once it connects. Cursor launches the server with npx, so Node.js must be installed.

Things you can now ask

In Cursor's Agent chat, the model calls the tools itself:

Summarize: "Summarize youtu.be/… — five takeaways with timestamps."
Pull quotes: "From this talk, list every quote about pricing, with timestamps."
Turn a video into docs: "Watch this setup walkthrough and draft the equivalent steps as a README."
Compare: "Search YouTube for three videos on Postgres indexing and summarize where they disagree."

That last one chains tools: search_videos to find candidates, get_video_metadata to triage, then extract_transcript on the keepers.

Why it's useful in an editor

The transcript lands right where you're working. Watched a conference talk about a library you're adopting? Ask Cursor to pull the transcript and turn the relevant part into setup steps or a code comment — no window-switching, no copy-paste. Because every segment is timestamped, Cursor can cite "at 14:20 they recommend…", so you can jump back to the source.

Keeping it cheap

search_videos, get_video_metadata, and get_related_videos are free, and extractions are cached — re-reading a video your agent already pulled costs nothing. Explore a topic across many videos and you only pay for the transcripts you actually extract. See pricing.

Frequently asked questions

Is this the same server Claude and Windsurf use?

Yes — one package, scribefy-mcp, works in any MCP host over stdio. See the Claude and Windsurf setups, or the overview for all assistants.

The tools don't show up — what now?

Make sure Node.js is installed (Cursor calls npx), the scribefy server is enabled in Settings → MCP, and your SCRIBEFY_API_KEY is set. Restart Cursor if the green dot doesn't appear.

What if a video has no captions?

Extraction will report that no captions are available — Scribefy reads captions (human or auto-generated), it doesn't run its own speech-to-text. See transcripts without captions.

Ready to wire it up? Grab a key from the dashboard, drop the scribefy block into mcp.json, and ask Cursor to read its first video.