voicevox-mcp

This project is an MCP (Model Context Protocol) server that can synthesize speech and obtain speaker information in cooperation with the VOICEVOX engine. It is implemented in TypeScript and uses the MCP SDK.

function

• Get speaker information for the VOICEVOX engine (/speakers)

• Synthesize text to speech on a specified speaker and play it locally (/speak)

  • Mac only

set up

Starting the VOICEVOX engine (Docker recommended)

docker compose up -d

This will start the VOICEVOX engine on localhost:50021.

Related MCP server: AivisSpeech MCP Server

Install and build dependencies

npm install
npm run build 

How to use

Cursor setting example

{
  "mcpServers": {
    "voicevox-mcp": {
      "command": "node",
      "args": ["${Path to Repository}/dist/index.js"],
      "env": {
        "SPEAKER_ID": 8,
        "SPEED_SCALE": 1.2,
        "VOICEVOX_API_URL": "http://localhost:50021" 
      }
    }
  }
}

Set VOICEVOX_API_URL as needed.

• You can get a list of speakers from your MCP client using the speakers tool.

• The speak tool can synthesize text to speech and play it back locally (Mac is recommended as it uses the afplay command).

Main Dependencies

• @modelcontextprotocol/sdk

• zod

• typescript

Precautions

• Future improvements

  • Speech synthesis will not be available unless the VOICEVOX engine is running on localhost:50021.

  • If you are using an environment other than Mac, please change the afplay part accordingly.

license

MIT License