# format37/youtube_mcp [Health: Active]

**Category:** 🎙️ Speech-to-Text  
**Repository:** https://github.com/format37/youtube_mcp  
**GitHub Stars:** 32  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/format37-youtube-mcp

## Description
MCP server that transcribes YouTube videos to text. Uses yt-dlp to download audio and OpenAI's Whisper-1 for more precise transcription than youtube captions. Provide a YouTube URL and get back the full transcript splitted by chunks for long videos.

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What format37/youtube_mcp MCP server does

format37/youtube_mcp MCP server turns YouTube video audio into text for MCP clients. The workflow starts with a YouTube URL, downloads the audio through yt-dlp, and sends it to OpenAI's Whisper-1 model for transcription. The result is a full transcript divided into chunks, which is useful for longer videos that may not fit in one response.

This server is focused on transcription rather than general YouTube management.

## How it works

The server runs as a containerized service and exposes an HTTP endpoint using server-sent events at `/sse`. An MCP client connects to that endpoint through `mcp-remote`, then makes the transcription capability available to the assistant. The client supplies a YouTube URL when requesting a transcript.

Audio retrieval depends on yt-dlp and the provided YouTube cookies. After the audio is downloaded, the server uses the configured OpenAI API key to call Whisper-1. Long results are split into multiple transcript chunks instead of being returned as one large block.

## Setup and configuration

To install format37/youtube_mcp MCP server, clone the repository and create its environment file. The documented configuration includes these values:

- `CONTAINER_NAME`: container name, shown as `youtube_mcp_main` in the example.
- `PORT`: service port, shown as `7001` in the example.
- `MCP_KEY`: access key generated with `token_generator.py`.
- `OPENAI_API_KEY`: credential for OpenAI Whisper-1.

You must also extract YouTube cookies and place the resulting `cookies.txt` file in the repository's `mcp/` directory. The README points to `cookies.md` for the extraction procedure. Before starting the service, make the provided `compose.sh`, `logs.sh`, and `update.sh` scripts executable, then run `compose.sh`. The configured port must accept incoming connections.

For Claude Desktop, add an `mcp-remote` entry targeting `http://localhost:7001/sse` and pass the MCP authorization header. Restart the client after saving its configuration.

## Tools and capabilities

- Accepts a YouTube URL for transcription.
- Downloads YouTube audio with yt-dlp.
- Uses OpenAI Whisper-1 for transcription.
- Returns full transcripts in chunks for long videos.
- Exposes the service through an MCP SSE endpoint.

## Limitations and notes

format37/youtube_mcp MCP server requires both an OpenAI API key and YouTube cookies. Transcription quality and availability depend on successful audio retrieval and access to the OpenAI API. Its documented client setup uses Claude Desktop with `mcp-remote`; other client configurations are not described.

_Full upstream README: https://allmcps.com/mcp/format37-youtube-mcp/readme_