Official (Vendor)Active· 25dby ElevenLabs

ElevenLabs

Official ElevenLabs MCP server. Text-to-speech, voice cloning, transcription, and audio sound design from inside Claude Desktop, Cursor, Windsurf, and other MCP clients.

View repo Homepage

Install

uvx elevenlabs-mcp

Paste into your host's MCP config:

{
  "mcpServers": {
    "ElevenLabs": {
      "command": "uvx",
      "args": ["elevenlabs-mcp"],
      "env": {
        "ELEVENLABS_API_KEY": "<insert-your-api-key-here>"
      }
    }
  }
}

Env vars needed: ELEVENLABS_API_KEY

Review

ElevenLabs ships the official MCP for voice. Connecting it gives whatever client you're using (Claude Desktop, Cursor, Windsurf, anything that speaks MCP) the same set of capabilities you'd hit through ElevenLabs's web app: text-to-speech in dozens of voices, voice cloning, transcription with speaker identification, audio effects, sound design, and style transfer between recordings. Install is `uvx elevenlabs-mcp` if you're set up with uv, or `pip install elevenlabs-mcp` for a standard Python environment. The config snippet for Claude Desktop drops into `claude_desktop_config.json` and needs your `ELEVENLABS_API_KEY` from the dashboard. Fair warning: every action costs ElevenLabs credits from your subscription, so it's not the kind of MCP you fire off speculatively. The natural workflow is generating spoken content from a draft. You're writing a script, an article, or a video brief in your editor. You ask the agent to "narrate this in the Adam voice, save to ~/audio/draft.mp3" and a few seconds later it's there. For solo creators doing voice work, that loop is meaningfully faster than browser-tabbing back to the ElevenLabs UI, pasting text, downloading. Voice cloning is the standout. If you've trained your own voice (or have permission to use someone else's), you can have the agent produce narration in that voice as part of a longer pipeline: research, outline, script, narration, cuts. Combined with a video editor MCP this gets close to end-to-end content generation from a single chat. The thing to watch is credit burn. Voice clone calls and longer narrations chew through ElevenLabs credits quickly, and there's no built-in dry-run preview in the MCP. Set hard limits in the ElevenLabs dashboard if you're letting agents loose without oversight. For anyone already paying for ElevenLabs, the MCP is the right way to use it from inside an agent workflow. If you're not, the MCP doesn't change the value equation, you're still buying a TTS service.

Verdict

The official MCP every ElevenLabs subscriber should connect day one. Costs credits per use, so manage agent permissions accordingly.

Good at

First-party support direct from ElevenLabs, not a community wrapper
Full feature surface: TTS, cloning, transcription, sound design, style transfer
Standard `uvx` install, no Docker, no infrastructure
Works with Claude Desktop, Cursor, Windsurf, Codex, and other MCP clients
Pairs naturally with file system and video MCPs for end-to-end content pipelines

Watch out

Burns ElevenLabs credits on every call, no preview or dry-run
Requires existing paid ElevenLabs account, no free tier path
Python-only install (no native Node package), uv recommended
Voice cloning needs the source voice already trained inside ElevenLabs
No built-in budget cap, set limits in the ElevenLabs dashboard

Use cases

Generate narration for video and prototype work
Clone a voice for character or branded content
Transcribe audio with speaker identification
Convert recordings between different voice styles
Design soundscapes and audio environments

Getting started

## Get an API key Log into elevenlabs.io and grab an API key from your profile. Free-tier keys work but credit limits are low. ## Install the server ``` uvx elevenlabs-mcp ``` Or for non-uv setups: ``` pip install elevenlabs-mcp ``` ## Configure Claude Desktop Add to `claude_desktop_config.json`: ```json { "mcpServers": { "ElevenLabs": { "command": "uvx", "args": ["elevenlabs-mcp"], "env": { "ELEVENLABS_API_KEY": "<your-key>" } } } } ``` Restart Claude Desktop. The tools should appear in the MCP menu. ## Test it Ask: "Use ElevenLabs to read this sentence aloud in the Adam voice, save to ~/test.mp3." Check the file plays.

Works with

Claude DesktopClaude CodeCursorWindsurfCodexOther