Transcript MCP server for Claude, Cursor and any AI agent
Connect Transkrio once and your assistant can read any YouTube or TikTok video, Instagram Reel or podcast episode you paste. "Summarize this talk", "pull the five best quotes from this episode", "what tools does this tutorial use?"
Setup
Get a free API key (100 credits), then add the server to your client.
Claude Code
claude mcp add --transport http transkrio https://api.transkrio.com/mcp \ --header "x-api-key: YOUR_KEY"
Cursor, VS Code, Windsurf and other JSON-configured clients
{
"mcpServers": {
"transkrio": {
"url": "https://api.transkrio.com/mcp",
"headers": { "x-api-key": "YOUR_KEY" }
}
}
}
Claude.ai and Claude Desktop
Settings, Connectors, Add custom connector. URL https://api.transkrio.com/mcp, then add the request header x-api-key: YOUR_KEY.
Tools
| Tool | What it does |
|---|---|
get_transcript | Transcript of a YouTube, TikTok or Instagram video or a podcast episode, as plain text (optionally with [mm:ss] timestamps), SRT, VTT or JSON, in the language you ask for. |
get_video_metadata | Title, channel or show, duration, views and available caption languages. |
list_videos | Videos of a YouTube playlist or channel, or episodes of a podcast, newest first. |
Prompts that work well
- "Summarize https://youtu.be/... in 10 bullet points with timestamps."
- "List the last 5 videos of @fireship and tell me which ones are about databases."
- "Get the transcript of this podcast episode and write show notes with chapters."
- "Translate the subtitles of this video to Spanish and give me an SRT file."
Why an MCP server instead of browsing
Assistants that browse can't read YouTube transcripts: the page loads them with JavaScript and YouTube blocks most automated requests. Through MCP the model gets the exact transcript as text, in one tool call, with timestamps it can cite. Building your own agent instead? Use the REST API directly, or the n8n node as a tool for n8n's AI Agent.