MCP.so
Sign In

MLX Whisper MCP Server

@kachiO

About MLX Whisper MCP Server

Local MCP server for MLX Whisper transcription

Basic information

Category

Other

License

Apache-2.0

Runtime

python

Transports

stdio

Publisher

kachiO

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "mlx-whisper-mcp": {
      "command": "uv",
      "args": [
        "run",
        "mlx_whisper_mcp.py"
      ]
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is MLX Whisper MCP Server?

A Model Context Protocol server that provides audio transcription capabilities using MLX Whisper on Apple Silicon Macs. It can transcribe audio files, base64-encoded audio data, and YouTube videos.

How to use MLX Whisper MCP Server?

Run directly with uv run mlx_whisper_mcp.py. For Claude Desktop, use uv tool run fastmcp install mlx_whisper_mcp.py or manually configure the server in the Claude Desktop config JSON with the uv command and the script's working directory.

Key features of MLX Whisper MCP Server

  • Transcribes audio files directly from disk
  • Transcribes audio from base64-encoded data
  • Downloads and transcribes YouTube videos
  • Uses the high-quality mlx-community/whisper-large-v3-turbo model
  • Self-contained script with automatic dependency management via uv run
  • Saves transcription text files alongside audio files

Use cases of MLX Whisper MCP Server

  • Transcribe a local audio recording for note-taking
  • Translate a foreign-language audio file to English
  • Get a transcript of a YouTube video’s speech
  • Process audio clips sent as base64 data from other apps

FAQ from MLX Whisper MCP Server

What hardware does it require?

An Apple Silicon Mac (M-series) running Python 3.12 or higher, with uv installed.

How do I install dependencies?

Dependencies are handled automatically by uv run when starting the server. The Whisper model (≈1.6 GB) is downloaded on first run.

Where are downloaded YouTube videos stored?

They are saved to ~/.mlx-whisper-mcp/downloads unless the keep_file option is set to True.

Can I transcribe very long audio files?

The large model may cause memory pressure on very long audio files.

How does authentication work for YouTube downloads?

Some videos may be restricted or require authentication, which is not supported by the server.

Comments

More Other MCP servers