elevenlabs-mcp

MCP server for the ElevenLabs text-to-speech API. Provides tools that any MCP-compatible client (Claude Desktop, Claude Code, etc.) can use to generate speech from text.

Setup

cd elevenlabs-mcp
npm install

Set your API key:

export ELEVENLABS_API_KEY="your-key-here"

Test the server starts correctly:

node index.js
# Should print "ElevenLabs MCP server running" to stderr

Claude Desktop configuration

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "elevenlabs": {
      "command": "node",
      "args": ["/ABSOLUTE/PATH/elevenlabs-mcp/index.js"],
      "env": { "ELEVENLABS_API_KEY": "your-key-here" }
    }
  }
}

Replace /ABSOLUTE/PATH/ with the actual path to this directory.

Quick-start workflow

1. Find a voice — call list_voices to get available voice_id values
2. Generate speech — call text_to_speech with the voice_id and your text
3. Check quota — call get_quota to verify character consumption

Tools

text_to_speech (primary)

Converts text to an mp3 file. Max 5000 characters per call.

Params: text, voice_id, output_path, model_id, stability, similarity_boost, style, language_code

text_to_speech_long

For texts over 5000 characters. Splits into chunks and saves numbered files.

Params: text, voice_id, output_dir, base_filename, model_id, stability, similarity_boost

list_voices

Lists all voices in your account with voice_id, name, category, and preview URL.

list_models

Lists TTS models with capabilities and language support.

get_quota

Shows character usage, limit, remaining quota, and next reset date.

Common voice IDs (reference)

These are default ElevenLabs voices — always verify with list_voices as IDs can vary by account:

Voice	ID
Rachel	21m00Tcm4TlvDq8ikWAM
Bella	EXAVITQu4vr4xnSDxMaL

Recommended models

Model	Use case
eleven_multilingual_v2	Best quality, 32 languages
eleven_flash_v2_5	Ultra-low latency (~75ms), real-time
eleven_v3	Newest, most expressive

Notes

• Billing is per character consumed
• Use get_quota to monitor usage before large jobs
• Character count is logged to stderr on every TTS call
• text_to_speech_long splits by paragraph then sentence boundaries