Windsurf's Cascade agent can do more than edit code — give it the Scribefy MCP server and it can search YouTube, inspect videos, and pull full timestamped transcripts itself. Here's the setup and what it unlocks.

What you get

The Model Context Protocol (MCP) lets Cascade 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 — the "Up next" feed

Only extract_transcript spends credits; the other three are free, and cached extractions are free too.

Setup

You'll need a Scribefy API key from the dashboard (the API + MCP plan). Then add the server to Windsurf's MCP config.

Edit the config file. Open ~/.codeium/windsurf/mcp_config.json and add:

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

Or use the UI. In the Cascade panel, click the MCPs icon (top-right), or go to Windsurf Settings → Cascade → MCP Servers. You can add servers from the marketplace, or choose View raw config to edit mcp_config.json directly — paste the scribefy block there.

After saving, click Refresh in the MCP panel so Cascade reloads the server. It runs via npx, so Node.js must be installed. Once connected, the four Scribefy tools appear in Cascade.

Things you can now ask

In Cascade, just describe what you want — it calls the tools itself:

Summarize: "Summarize youtu.be/… in five bullet points with timestamps."
Extract steps: "This video sets up CI — turn it into a checklist I can follow."
Research: "Find three videos on gRPC vs REST and summarize the trade-offs they mention."
Quote-find: "Pull every timestamped quote about latency from this talk."

For research prompts, Cascade chains the tools: search_videos → get_video_metadata to triage → extract_transcript on the ones worth reading.

Why timestamps matter

Every Scribefy segment carries a timestamp, so Cascade can cite where something was said ("at 8:15 they benchmark it") and you can jump straight there. That turns a vague "the video says X" into a verifiable, source-linked answer.

Keeping it cheap

Search, metadata, and related-videos are free; extraction is cached. Exploring a topic across many videos stays cheap because you only pay for the transcripts you actually pull, once. See pricing.

Frequently asked questions

Is this the same server Claude and Cursor use?

Yes — scribefy-mcp is one package for any stdio MCP host. See the Claude and Cursor guides, or the overview for all assistants.

The tools aren't showing up

Confirm Node.js is installed (Cascade calls npx), your SCRIBEFY_API_KEY is set in mcp_config.json, and you clicked Refresh in the MCP panel. Reopen Windsurf if needed.

Does it transcribe videos that have no captions?

No — Scribefy reads existing captions (human or auto-generated). For caption-less videos, see transcripts without captions.

Ready? Grab a key from the dashboard, add the scribefy block to mcp_config.json, hit Refresh, and ask Cascade to read a video.