mcp-tts

mcp-tts

MCP server that gives Claude Code a voice via ElevenLabs TTS. Includes 13 voices, configurable voice modes, per-category voice selection, and smart summarization.

Setup

1. Install

Via plugin marketplace:

/plugin install mcp-tts@stitts-plugins

Or via npx (no clone needed):

{
  "tts": {
    "command": "npx",
    "args": ["-y", "github:stitts-dev/mcp-tts"],
    "env": {
      "ELEVENLABS_API_KEY": "your-api-key"
    }
  }
}

Or git clone:

cd ~/.claude
git clone https://github.com/stitts-dev/mcp-tts.git mcp-tts
cd mcp-tts && npm install

2. Environment variables

Variable	Required	Default	Description
ELEVENLABS_API_KEY	Yes	—	Your ElevenLabs API key
ELEVENLABS_VOICE_ID	No	edward	Override default voice by ID
ELEVENLABS_MODEL_ID	No	eleven_multilingual_v2	ElevenLabs model

3. Register MCP server

Add to ~/.claude.json (or ~/.claude/.mcp.json) inside "mcpServers":

"tts": {
  "type": "stdio",
  "command": "node",
  "args": ["~/.claude/mcp-tts/index.mjs"],
  "env": {
    "ELEVENLABS_API_KEY": "your-api-key"
  }
}

4. Add instructions to CLAUDE.md

Copy INSTRUCTIONS.md contents into your ~/.claude/CLAUDE.md. This tells Claude when and how to use each voice.

Tools

speak

Parameter	Type	Required	Description
text	string	Yes	Text to speak (markdown stripped automatically)
voice	string	No	Voice override (server auto-selects from config)
category	string	No	Response category: completions, errors, questions, status, summaries

The server decides whether to speak based on voice config mode and category settings. Rate-limited to one speak every 3 seconds.

get_voice_config

Returns current voice configuration as JSON. No parameters.

set_voice_config

Parameter	Type	Required	Description
config	object	Yes	Full voice config object (replaces entire config)

Config is stored at ~/.claude/voice-config.json (persists across reinstalls).

Voice Modes

Configure via /tts skill or set_voice_config:

Mode	What speaks
off	Nothing
minimal	Task completions, build failures, blocking questions (default)
ambient	All of minimal + status updates + long-output summaries
full	Everything — every response gets spoken (summaries for long output)

Available voices

Voice	Tone	Best for
edward	British, deep, composed	Default — task completions, neutral updates
daniel	Steady broadcaster	Factual summaries, status reports
donovan	Articulate, strong	Confident recommendations, warnings
archer	Conversational	Casual updates
juniper	Grounded, professional	Professional summaries
cassidy	Crisp, direct	Short alerts
austin	Deep, raspy, authentic	Informal assessments
mark-natural	Young, natural	Casual conversation
mark-casual	Relaxed, light	Low-stakes updates
jessica	Eloquent, dramatic	Impressive results
spuds	Wise, approachable	Advice, mentoring
rob	Tough, gritty	Bad news, hard truths
adam	Dark, tough	Critical failures

Test

export ELEVENLABS_API_KEY=sk_...
cd ~/.claude/mcp-tts && node index.mjs --test

Requirements

• macOS (uses afplay for audio playback)
• Node.js 18+
• ElevenLabs API key (elevenlabs.io)