MCP Youtube Transcript

What mcp-youtube-transcript does

mcp-youtube-transcript gives MCP-compatible agents access to transcript and metadata retrieval for YouTube videos. Agents provide a full YouTube video URL, then select the operation they need:

• Retrieve transcript text.
• Retrieve transcript segments with timing information.
• List manual and automatically generated transcript languages.
• Fetch video metadata such as its title and channel.

Transcript requests accept an optional language code and default to English. The server is useful for workflows that need to summarize lectures, analyze interviews, extract quotations, or inspect the contents of long-form videos without manually opening them.

How it works

The server communicates over standard MCP stdio. A client starts it with uvx, using the Git repository as the package source, and sends tool calls containing a YouTube URL. Transcript tools return responses in chunks when they exceed the configured character limit. The default limit is 50,000 characters; a returned cursor lets an agent request the next portion with next_cursor.

The response limit can be changed with --response-limit. Passing a negative value disables pagination and requests the complete transcript in one response. Timestamped results include segment start times and durations, which supports time-aware navigation and citations.

Setup and configuration

Install Astral uv to use the uvx command, then add an MCP server entry similar to this configuration:

{
  "mcpServers": {
    "youtube-transcript": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/jkawamoto/mcp-youtube-transcript",
        "mcp-youtube-transcript"
      ]
    }
  }
}

Claude Desktop also offers an .mcpb bundle from the project's releases, while Cursor and LM Studio provide their own MCP configuration flows. A Docker image is available through Docker Hub.

For proxy routing, configure WEBSHARE_PROXY_USERNAME and WEBSHARE_PROXY_PASSWORD for Webshare, SCRAPINGANT_API_TOKEN for ScrapingAnt, or HTTP_PROXY and HTTPS_PROXY for generic proxy endpoints. The same settings can be supplied through command-line flags.

Tools and capabilities

get_transcript accepts a required video URL, an optional language code, and an optional pagination cursor. get_timed_transcript uses the same inputs and returns timed segments. get_available_languages lists the transcript languages associated with a video, and get_video_info returns video information. The language and metadata tools require only the video URL.

Limitations and notes

Long responses may be split into multiple calls unless pagination is disabled or the response limit is increased. YouTube may block automated requests from data-center or cloud IP addresses, so proxy configuration can be necessary for reliable access. ScrapingAnt access requires a paid subscription, and residential proxy usage consumes more credits than standard proxy requests. The project is licensed under MIT.