trawl
Back to Playbooks
AI/ML Engineereasy
mcpclaudeai-agentstutorial

Giving Claude Access to YouTube Transcripts via MCP

Trawl Team·

I wanted Claude to be able to look up YouTube transcripts mid-conversation without me copying URLs back and forth. The workflow was always the same: I'd ask Claude a question, realize I needed context from a video, open a new tab, find the video, copy the URL, paste it into Trawl, copy the transcript, paste it back into Claude. Five steps that should be zero.

The Model Context Protocol (MCP) eliminates all of that. With Trawl running as an MCP server, Claude can autonomously search YouTube, extract transcripts, and feed them into its analysis — all within the conversation.

Why this matters

The MCP ecosystem has grown to 97 million monthly SDK downloads. Anthropic, OpenAI, Google DeepMind, and Microsoft have all adopted the protocol. Running Trawl as an MCP server makes transcript extraction a first-class capability in any MCP-compatible agent.

The key insight: instead of you being the middleman between Claude and video content, Claude reaches for transcripts the same way it reaches for any other tool.

Setup in 60 seconds

Add this to your Claude Desktop config (~/.config/claude/claude_desktop_config.json):

{
  "mcpServers": {
    "trawl": {
      "command": "python",
      "args": ["/path/to/mcp-server/server.py"],
      "env": {
        "BACKEND_URL": "https://api.gettrawl.com",
        "TRAWL_API_KEY": "trawl_your_api_key"
      }
    }
  }
}

Restart Claude Desktop. You'll see four new tools available.

Available tools

ToolWhat it does
get_transcriptExtract a transcript from any YouTube URL
search_videosSearch YouTube by keyword
list_historyBrowse your extraction history
get_transcript_by_idRetrieve a previously extracted transcript

What Claude actually returns

Here's a real example. I asked Claude: "Search YouTube for 'FOMC press conference March 2026' and summarize the key policy changes."

Claude used search_videos to find the press conference, called get_transcript on the top result, and returned something like:

The March 2026 FOMC press conference covered three main policy changes:

  1. Rate hold at 4.25-4.50% — Chair Powell emphasized that the committee needs "more evidence of sustained disinflation" before cutting further. [timestamp: 03:42]

  2. QT pace reduction — Treasury runoff cap reduced from $25B to $5B/month, effective April. [timestamp: 18:15]

  3. Updated dot plot — Median projection shifted from 3 cuts to 2 cuts for 2026. [timestamp: 24:30]

No tab-switching. No copy-pasting. Claude found the video, read the transcript, and synthesized the answer in one turn.

Try the API Yourself

Before setting up MCP, try the same endpoints Claude will call under the hood:

Extract transcript
curl -X POST "https://api.gettrawl.com/api/transcripts/preview" \
  -H "Content-Type: application/json" \
  -d '{
  "url": "https://www.youtube.com/watch?v=8jPQjjsBbIc"
}'

Example prompts

Once connected, try these:

  • "Search YouTube for 'FOMC press conference March 2026' and extract the transcript of the top result."
  • "Get the transcript of https://youtube.com/watch?v=abc123 and summarize the key points."
  • "Search for 'machine learning interview questions', extract transcripts from the top 3 results, and create a study guide."

How it works technically

The MCP server is a standalone Python process that communicates with Claude via stdio transport. When Claude calls a tool, the server makes an authenticated API call to Trawl's backend and returns the result. The server reads TRAWL_API_KEY from the environment. All operations count toward your account's usage quota.

The 78 tools in the full MCP server cover every Trawl source — YouTube, podcasts, TikTok, Instagram, Reddit, earnings calls, SEC filings, news, and more. The Claude Desktop config above gives you access to all of them.

TwitterLinkedIn