Voice Multi-Provider TTS System

Overview

Voice MCP Server v5.0 implements flexible TTS provider switching between ElevenLabs and Inworld AI.

Cost Comparison

ElevenLabs: $100/month subscription
Inworld AI: FREE until Dec 2025, then $5/1M characters (95% savings)

Configuration

File: /opt/mcp-servers/voice/mcp_voice_server.py

VOICE_CONFIG = {
    "default": {
        "provider": "inworld",  # or "elevenlabs"
        "inworld_voice_id": "default-r3uenniwomhoqljb1p7ysg__lena",
        "elevenlabs_voice_id": "cgSgspJ2msm6clMCkdW9",
    },
    "lars": {
        "provider": "inworld",
        "inworld_voice_id": "Edward",
        "elevenlabs_voice_id": "UgBBYS2sOqTuMpoF3BR0",
    },
}

Quick Switch

To switch a voice back to ElevenLabs: 1. Open /opt/mcp-servers/voice/mcp_voice_server.py 2. Change "provider": "inworld" to "provider": "elevenlabs" 3. Gateway auto-reloads on next call

Voice IDs

Inworld

Lena (cloned): default-r3uenniwomhoqljb1p7ysg__lena
Lars (Edward): Edward (built-in)
Other built-in voices: Ashley, etc.

ElevenLabs (fallback)

Lena: cgSgspJ2msm6clMCkdW9
Lars: UgBBYS2sOqTuMpoF3BR0

Inworld API

Endpoint: https://api.inworld.ai/tts/v1/voice
Auth: Basic (base64 API key)
Model: inworld-tts-1
Credentials: Locker l_684a

Pronunciation Tips

Use proper case for names: 'Lars' not 'LARS'
ALL CAPS interpreted as acronyms and spelled out

KB: b4e6ab28 (Inworld research)
Locker: l_684a (Inworld credentials)