A Model Context Protocol server that provides text-to-speech capabilities using the Kokoro TTS model.
The server can be configured using the following environment variables:
Variable | Description | Default | Valid Range |
---|---|---|---|
MCP_DEFAULT_SPEECH_SPEED |
Default speed multiplier for text-to-speech | 1.1 | 0.5 to 2.0 |
MCP_DEFAULT_VOICE |
Default voice for text-to-speech | af_bella | Any valid voice ID |
In Cursor:
{
"mcpServers": {
"speech": {
"command": "npx",
"args": [
"-y",
"speech-mcp-server"
],
"env": {
"MCP_DEFAULT_SPEECH_SPEED": 1.3,
"MCP_DEFAULT_VOICE": "af_bella"
}
}
}
}
- 🎯 High-quality text-to-speech using Kokoro TTS model
- 🗣️ Multiple voice options available
- 🎛️ Customizable speech parameters (voice, speed)
- 🔌 MCP-compliant interface
- 📦 Easy installation and setup
- 🚀 No API key required
# Using npm
npm install speech-mcp-server
# Using pnpm (recommended)
pnpm add speech-mcp-server
# Using yarn
yarn add speech-mcp-server
Run the server:
# Using default configuration
npm start
# With custom configuration
MCP_DEFAULT_SPEECH_SPEED=1.5 MCP_DEFAULT_VOICE=af_bella npm start
The server provides the following MCP tools:
text_to_speech
: Basic text-to-speech conversiontext_to_speech_with_options
: Text-to-speech with customizable speedlist_voices
: List all available voicesget_model_status
: Check the initialization status of the TTS model
# Clone the repository
git clone <your-repo-url>
cd speech-mcp-server
# Install dependencies
pnpm install
# Start development server with auto-reload
pnpm dev
# Build the project
pnpm build
# Run linting
pnpm lint
# Format code
pnpm format
# Test with MCP Inspector
pnpm inspector
Converts text to speech using the default settings.
{
"type": "request",
"id": "1",
"method": "call_tool",
"params": {
"name": "text_to_speech",
"arguments": {
"text": "Hello world",
"voice": "af_bella" // optional
}
}
}
Converts text to speech with customizable parameters.
{
"type": "request",
"id": "1",
"method": "call_tool",
"params": {
"name": "text_to_speech_with_options",
"arguments": {
"text": "Hello world",
"voice": "af_bella", // optional
"speed": 1.0, // optional (0.5 to 2.0)
}
}
}
Lists all available voices for text-to-speech.
{
"type": "request",
"id": "1",
"method": "list_voices",
"params": {}
}
Check the current status of the TTS model initialization. This is particularly useful when first starting the server, as the model needs to be downloaded and initialized.
{
"type": "request",
"id": "1",
"method": "call_tool",
"params": {
"name": "get_model_status",
"arguments": {}
}
}
Response example:
{
"content": [{
"type": "text",
"text": "Model status: initializing (5s elapsed)"
}]
}
Possible status values:
uninitialized
: Model initialization hasn't startedinitializing
: Model is being downloaded and initializedready
: Model is ready to useerror
: An error occurred during initialization
You can test the server using the MCP Inspector or by sending raw JSON messages:
# List available tools
echo '{"type":"request","id":"1","method":"list_tools","params":{}}' | node dist/index.js
# List available voices
echo '{"type":"request","id":"2","method":"list_voices","params":{}}' | node dist/index.js
# Convert text to speech
echo '{"type":"request","id":"3","method":"call_tool","params":{"name":"text_to_speech","arguments":{"text":"Hello world","voice":"af_bella"}}}' | node dist/index.js
To use this server with Claude Desktop, add the following to your Claude Desktop config file (~/Library/Application Support/Claude/claude_desktop_config.json
):
{
"servers": {
"speech": {
"command": "npx",
"args": ["@decodershq/speech-mcp-server"]
}
}
}
Contributions are welcome! Please feel free to submit a Pull Request.
MIT License - see the LICENSE file for details.
The server automatically attempts to download and initialize the TTS model on startup. If you encounter initialization errors:
- The server will automatically retry up to 3 times with a cleanup between attempts
- Use the
get_model_status
tool to monitor initialization progress and any errors - If initialization fails after all retries, try manually removing the model files:
# Remove model files (MacOS/Linux)
rm -rf ~/.npm/_npx/**/node_modules/@huggingface/transformers/.cache/onnx-community/Kokoro-82M-v1.0-ONNX/onnx/model_quantized.onnx
rm -rf ~/.cache/huggingface/transformers/onnx-community/Kokoro-82M-v1.0-ONNX/onnx/model_quantized.onnx
# Then restart the server
npm start
The get_model_status
tool will now include retry information in its response:
{
"content": [{
"type": "text",
"text": "Model status: initializing (5s elapsed, retry 1/3)"
}]
}