YouTube Transcript MCP

YouTube Transcript MCP Server for Claude Code, Cursor and Windsurf

TubeTranscript is a hosted MCP server that lets AI agents fetch the transcript of a YouTube video, including videos without captions. Pro API key required.

This repository holds setup configs, examples and docs only. The server itself is hosted and closed-source, and every tool call needs a Pro API key.

Quick start

TubeTranscript runs two remote MCP servers. https://tubetranscript.io/api/mcp fetches transcripts, and https://tubetranscript.io/api/mcp/tools turns a transcript into chapters, hooks, a mindmap, subtitles or a blog draft. Both take the same Pro API key in an Authorization: Bearer header. The AI tools take the job_id that get_transcript returns, so register both.

In the examples, replace the placeholder tt_live_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX with your key, and keep any file that contains the key out of version control.

Claude Code

Export your key as TUBETRANSCRIPT_API_KEY, then register both servers:

claude mcp add --transport http tubetranscript https://tubetranscript.io/api/mcp \
  --header "Authorization: Bearer $TUBETRANSCRIPT_API_KEY"
claude mcp add --transport http tubetranscript-ai https://tubetranscript.io/api/mcp/tools \
  --header "Authorization: Bearer $TUBETRANSCRIPT_API_KEY"

Run claude mcp list to confirm both connect, then start a new session so the tools load. More in examples/claude-code.

Cursor

Add both servers to .cursor/mcp.json in the project, or to ~/.cursor/mcp.json for every project. Cursor detects the Streamable HTTP transport from the URL. The file is also at examples/cursor/mcp.json.

{
  "mcpServers": {
    "tubetranscript": {
      "url": "https://tubetranscript.io/api/mcp",
      "headers": { "Authorization": "Bearer tt_live_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX" }
    },
    "tubetranscript-ai": {
      "url": "https://tubetranscript.io/api/mcp/tools",
      "headers": { "Authorization": "Bearer tt_live_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX" }
    }
  }
}

Windsurf

Add both servers to ~/.codeium/windsurf/mcp_config.json. Windsurf uses serverUrl for remote servers. After saving, refresh the MCP server list in Cascade. The file is also at examples/windsurf/mcp_config.json.

{
  "mcpServers": {
    "tubetranscript": {
      "serverUrl": "https://tubetranscript.io/api/mcp",
      "headers": { "Authorization": "Bearer tt_live_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX" }
    },
    "tubetranscript-ai": {
      "serverUrl": "https://tubetranscript.io/api/mcp/tools",
      "headers": { "Authorization": "Bearer tt_live_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX" }
    }
  }
}

n8n

Add an MCP Client Tool node to an AI Agent workflow and set:

Add a second MCP Client Tool node for /api/mcp/tools. See examples/n8n.

Give Claude Code a YouTube transcript

Once both servers are registered, ask in plain language:

Use get_transcript to fetch https://www.youtube.com/watch?v=jNQXAC9IVRw and summarize it in five bullet points. If the status is "processing", wait about a minute and call it again.

A typical agent workflow takes two or three calls:

1. The agent calls get_transcript with the video URL or ID and gets the transcript as plain text, plus a job_id.
2. If status is processing, the video has no captions and its audio is being transcribed. The agent calls again with the same video after about 60 seconds.
3. The agent passes the job_id to an AI tool such as generate_chapters, extract_hooks or translate_subtitles.

Repeat calls are safe: your team's existing job for a video is reused, so nothing is transcribed twice.

Tools

get_transcript accepts video_url (any YouTube URL) or video_id (the 11-character ID). The AI tools read the transcript of a job you already have and never fetch videos themselves. Calling one again with the same job_id returns the stored result. Full input and output fields are in the MCP setup guide.

YouTube videos without captions

When a video has no captions, get_transcript falls back to audio transcription instead of returning an error. It returns status: "processing" and an empty transcript; call it again with the same video after about a minute until the status is completed. On Pro it handles videos up to 2 hours. Private and removed videos, and live streams in progress, still fail.

Limits, authentication and errors

The 402 body carries a reason:

Arguments are validated before any quota is spent: a missing, mistyped or unknown argument returns HTTP 400 with JSON-RPC error -32602. A tool that runs but cannot finish returns HTTP 200 with result.isError: true and a JSON block { "error": "<CODE>", "message": "…" }, for example JOB_NOT_FOUND or JOB_NOT_READY, so always check isError. The full list is in the MCP setup guide. Key handling and limits: security and rate limits.

Use the REST API instead

If your own code, a script or an n8n HTTP Request node decides when to fetch a transcript, call the REST API with the same Pro key and the same daily limit. It returns timed chunks as JSON.

curl -X POST https://tubetranscript.io/api/v1/extract \
  -H "Authorization: Bearer $TUBETRANSCRIPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://www.youtube.com/watch?v=jNQXAC9IVRw"}'

Python and JavaScript versions that wait for a transcription in progress are in examples/rest-api.

Compared with open-source YouTube transcript MCP servers

Free open-source servers such as jkawamoto/mcp-youtube-transcript and kimtaeyoon83/mcp-server-youtube-transcript are a good fit when you are happy to run a process on your own machine. Read from their READMEs on 2026-10-16:

Choose an open-source server if you only need captions, want it free, or want timed segments inside MCP.

FAQ

What is the TubeTranscript MCP server? Two remote MCP servers that let an AI agent fetch the transcript of a YouTube video and turn it into hooks, chapters, a mindmap, subtitles or a blog draft. You register them once in Claude Code, Cursor, Windsurf or n8n, and the agent calls the tools itself.

How much does it cost? The MCP servers are part of the Pro plan, $9/month. One Pro API key makes up to 1,000 tool calls per 24 hours, shared with the REST API. See pricing.

Does get_transcript work for videos without captions? Yes. It reads the captions first. With none, it transcribes the audio and returns status processing; the agent calls it again with the same video after about a minute until the status is completed. Pro covers videos up to 2 hours.

Do I need Pro just to try it? For the MCP servers, yes: every tool call needs a Pro API key. To see what a transcript looks like first, paste a link on the home page, or use Copy for Claude, which needs no account.

Should I use MCP or the REST API? Use MCP when an agent decides when to fetch a transcript. Use the REST API when your own code does, for example from a script or an automation. Both use the same Pro key and the same daily limit.

How is this different from the free open-source YouTube transcript MCP servers? Those run on your machine and are free. This one is hosted, so there is no process or proxy to set up, and it falls back to audio transcription when a video has no captions. It costs Pro, and get_transcript returns plain text; timed segments come from the REST API. See the comparison above.

Which clients are supported? Any client that speaks Streamable HTTP MCP with a Bearer header. The setup guide documents Claude Code, Cursor, Windsurf and n8n.

Is the server source code in this repository? No. The server is hosted and closed-source. This repository has the setup configs, examples and registry metadata for it.

Links

• YouTube Transcript MCP Server: the product page
• MCP setup guide: every step, field and error code
• YouTube Transcript API: the REST endpoint
• Transcripts for videos without captions
• Pricing
• Registry metadata: registry/server.transcripts.json (io.tubetranscript/transcripts) and registry/server.ai-tools.json (io.tubetranscript/ai-tools)

TubeTranscript.io is not affiliated with tubetranscript.com. The examples and docs in this repository are MIT licensed. The hosted service is governed by https://tubetranscript.io/terms.