Interactive Voice

You can run an Interactive Voice MCP Server that combines Kokoro Text-to-Speech and NVIDIA NeMo Parakeet speech-to-text to enable natural, two-way voice conversations. You start the server locally and connect it to your MCP client to speak prompts and listen to user responses in real time.

How to use

You interact with the Interactive Voice MCP Server through an MCP client that supports the interactive_voice_dialog tool. When you send a text prompt, the server speaks the response aloud and then listens for your spoken reply. The tool returns a transcription of what you said, which your client then uses as the next input. Use a brief, friendly voice prompt summary to transition into listening, and then wait for the user’s spoken input to continue the conversation.

Key capabilities you can rely on include synthesizing speech from text, playing the audio for the user, and capturing the user’s spoken input for transcription. You can choose a voice for speech output with the voice parameter (for example, af_heart). If you don’t specify a voice, the system uses the default voice.

How to install

Prerequisites include having the espeak-ng tool installed on your system, since some underlying TTS models rely on it.

Windows installation steps for espeak-ng:

• Go to espeak-ng releases
• Click on "Latest release"
• Download the appropriate *.msi file (for example, espeak-ng-20191129-b702b03-x64.msi)
• Run the downloaded installer

Local development installation so you can run the server from Python during development: navigate to the directory containing the project’s build configuration (pyproject.toml) and install in editable mode so changes are reflected immediately.

pip install -e .

After installation, start the server using one of the following commands depending on your preferred module path. The recommended approach is shown here.

# Option 1: run as a module under the package name
python -m mcp_server_tts.server

# Option 2: run using a different package path if you adapt the structure
python -m mcp_interactive_voice_server

Configuration and usage notes

To connect from your MCP client, register the server under the mcpServers section in your Claude Desktop configuration.

Use the following example to configure Claude Desktop to launch the TTS server directly from Python.

{
  // ... other configurations ...
  "mcpServers": {
    // ... other servers ...
    "tts": {
      "command": "python",
      "args": ["-m", "mcp_server_tts"]
    }
    // ... other servers ...
  }
  // ... other configurations ...
}

Troubleshooting and notes

If you encounter issues with voice synthesis or transcription, verify that espeak-ng is installed and accessible on your path. Ensure your Python environment has the necessary dependencies installed in editable mode so changes to the source are picked up immediately during development.

Examples and security considerations

When deploying in a shared environment, keep your local MCP server accessible to your MCP client and avoid exposing the server beyond trusted networks. Use default or clearly named voices to help users understand the spoken output, and provide a simple way for users to adjust or disable voice input if needed.

Available tools

interactive_voice_dialog

Synthesizes text to speech, plays it, then listens for user speech input and returns the transcription. Requires text_to_speak and optional voice to select a TTS voice.