CosyVoice MCP Server

A Model Context Protocol (MCP) server for Alibaba Cloud's CosyVoice text-to-speech service.

This server allows you to generate high-quality, expressive speech from text using Alibaba Cloud's state-of-the-art CosyVoice models (V2 and V3).

Features

• List Voices: Retrieve a curated list of high-quality voices suitable for various scenarios (social, narration, customer service, etc.).
• Speech Synthesis: Convert text to speech with support for:
  • Model Selection: Choose between cosyvoice-v2 (standard) and cosyvoice-v3-flash (emotional).
  • Control Parameters: Adjust volume, speech rate, and pitch rate.
  • Emotion Control: Use natural language instructions to control emotion (V3 models only).
    • Supported emotions: neutral, fearful, angry, sad, surprised, happy, disgusted.
    • Format: "你说话的情感是<emotion>。"
  • SSML Support: Use SSML tags (e.g., <break time="2s"/>) for precise pause control.
  • File Saving: Save generated audio directly to a local file path.

Configuration

You need an Alibaba Cloud DashScope API Key. Set it as an environment variable:

export COSYVOICE_MCP_DASHSCOPE_API_KEY=your_api_key_here

(Alternatively, DASHSCOPE_API_KEY is also supported).

Optional Configuration

• COSYVOICE_DEFAULT_OUTPUT_DIR: Set a default directory for saving generated audio files if output_file is not specified.

export COSYVOICE_DEFAULT_OUTPUT_DIR=~/Desktop/CosyVoiceOutput

MCP Client Configuration

Add the following to your MCP client configuration (e.g., claude_desktop_config.json):

Using uvx (Recommended)

You can run the server directly from PyPI using uvx:

{
  "mcpServers": {
    "cosyvoice": {
      "command": "uvx",
      "args": [
        "voicemcpserver"
      ],
      "env": {
        "COSYVOICE_MCP_DASHSCOPE_API_KEY": "your_api_key_here"
      }
    }
  }
}

Using python

{
  "mcpServers": {
    "cosyvoice": {
      "command": "python3",
      "args": [
        "-m",
        "voicemcpserver"
      ],
      "cwd": "/path/to/voicemcpserver/src",
      "env": {
        "COSYVOICE_MCP_DASHSCOPE_API_KEY": "your_api_key_here"
      }
    }
  }
}

Available Voices

CosyVoice V3 Flash (Supports Emotion Instructions)

CosyVoice V2 (Standard High Quality)

Usage Examples

Synthesize Speech (Basic)

{
  "text": "你好，欢迎使用 CosyVoice。",
  "voice": "longxiaochun_v2",
  "output_file": "/Users/username/Desktop/hello.mp3"
}

Synthesize Speech (Advanced Control)

{
  "text": "今天天气真不错！<break time=\"1s\"/> 我们去公园玩吧。",
  "voice": "longanyang",
  "model": "cosyvoice-v3-flash",
  "volume": 80,
  "speech_rate": 1.1,
  "instruction": "你说话的情感是happy。",
  "output_file": "/Users/username/Desktop/happy.mp3"
}