Connect Fish Audio TTS to Voice Mode MCP

Voice Mode adds two-way voice conversations to Claude Code and other MCP clients. Add the Fish Audio OpenAI-compatible endpoint to its TTS service list.

Configure environment variables

Add these values to the MCP server's env block:

{
  "OPENAI_API_KEY": "YOUR_API_KEY",
  "VOICEMODE_TTS_BASE_URLS": "https://fishaudio.org/v1",
  "VOICEMODE_TTS_MODEL": "fishaudio-s21pro-flash",
  "VOICEMODE_TTS_VOICE": "00a1b221-6137-4b73-ad62-b0cbce134167"
}

Alternatively, put them in ~/.voicemode/voicemode.env:

export OPENAI_API_KEY="YOUR_API_KEY"
export VOICEMODE_TTS_BASE_URLS="https://fishaudio.org/v1"
export VOICEMODE_TTS_MODEL="fishaudio-s21pro-flash"
export VOICEMODE_TTS_VOICE="00a1b221-6137-4b73-ad62-b0cbce134167"

The legacy TTS_BASE_URL, TTS_MODEL, and TTS_VOICE variables remain supported, but new configurations should use the VOICEMODE_ names.

Test and change the voice

Restart the MCP client so the Voice Mode server reloads its environment. Start a short voice conversation, then replace the test voice with your preferred Voice ID.

When multiple TTS URLs are configured, Voice Mode can try fallback services in sequence. Keep only https://fishaudio.org/v1 while troubleshooting.

Troubleshooting

• Another service is used: remove other entries from VOICEMODE_TTS_BASE_URLS during testing.
• API key is missing: ensure the variable is available to the process that starts the MCP server.
• Invalid model or voice: use Fish Audio IDs rather than OpenAI defaults.
• 401/402: verify the key and API credit balance.

Related documentation

• API authentication
• OpenAI-Compatible TTS
• Voice Mode environment reference