ElevenLabs MCP Server: Generate Voice with Claude

Writing a blog post and want a podcast version? Producing a product demo video and need a professional voiceover? Normally that means copying your text, opening ElevenLabs in another tab, selecting a voice, downloading the file, and then resuming your work. The ElevenLabs MCP server eliminates every one of those steps. It lets Claude generate voice audio directly — you describe what you want, Claude calls ElevenLabs, and the MP3 file appears on your disk while you stay in the conversation.

What is ElevenLabs?

ElevenLabs is a leading AI audio platform specializing in highly realistic text-to-speech synthesis and voice cloning. Founded in 2022, it quickly became the go-to tool for content creators, podcasters, game developers, and enterprise teams who need natural-sounding AI voices. ElevenLabs offers a library of pre-made voices in dozens of languages and accents, plus the ability to clone custom voices from audio samples. Its models — including Multilingual v2 and Turbo — are widely regarded as among the most lifelike TTS systems available.

What the ElevenLabs MCP Server Does

The ElevenLabs MCP server exposes ElevenLabs' text-to-speech API as a set of tools Claude can call during your conversations. The core capabilities include:

• Generate speech from text — convert any text to audio, specifying voice ID, model (quality vs. speed), and output format (MP3, WAV, OGG).
• List available voices — retrieve all voices on your account, including pre-made library voices and any custom/cloned voices you've created.
• Get voice details — fetch metadata for a specific voice including its settings, labels, and sample audio URL.
• Convert text to a local audio file — the generated audio is saved directly to your filesystem with the filename you specify.

Claude decides which voice to use based on your description, constructs the API request, streams the audio, and saves it — all in a single step from your perspective. To understand how Claude reasons about which tool to call, see our MCP tool schema explained guide.

Prerequisites

Before setting up the ElevenLabs MCP server, you will need:

• ElevenLabs account — create a free account at elevenlabs.io. The free tier includes 10,000 characters per month.
• ElevenLabs API key — go to your Profile page (top-right avatar) → API Key. Copy the key — it starts with a long alphanumeric string.
• Node.js 18 or higher — verify with node --version. The server package is distributed via npm and launched with npx.
• Claude Desktop — MCP servers connect through the Claude Desktop app. See our MCP server installation guide if this is your first time.

How to Install the ElevenLabs MCP Server

1. Get your ElevenLabs API key. Log in at elevenlabs.io, click your avatar in the top right, select Profile, and copy the API key shown on that page.
2. Open claude_desktop_config.json. On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows: %APPDATA%\Claude\claude_desktop_config.json. Full details in our claude_desktop_config.json guide.
3. Add the ElevenLabs server block to the mcpServers object using the configuration below.
4. Paste your API key in place of your_api_key_here.
5. Save the file and restart Claude Desktop. The ElevenLabs server will be available immediately on next launch.

The Config JSON

Add this to the mcpServers section of your claude_desktop_config.json. The official ElevenLabs package is published and maintained by ElevenLabs themselves:

{
  "mcpServers": {
    "elevenlabs": {
      "command": "npx",
      "args": ["-y", "@elevenlabs/elevenlabs-mcp"],
      "env": {
        "ELEVENLABS_API_KEY": "your_api_key_here"
      }
    }
  }
}

This is one of the few MCP servers with an officially maintained package from the service provider itself, which means updates and compatibility are well-maintained. To understand best practices for storing API keys securely, see our MCP security guide.

Real Use Cases

Here are the kinds of voice generation tasks you can hand off to Claude once the server is connected:

1. "Read this blog post aloud in Rachel's voice and save it as podcast-intro.mp3." Claude takes the blog post text, calls the TTS API with Rachel's voice ID, and saves the output to podcast-intro.mp3 in your current directory.
2. "Generate a voiceover for this ad script using a British male voice." Claude selects an appropriate British male voice from your account library and produces a ready-to-use voiceover file.
3. "List all the voices available on my account." Claude retrieves and displays your full voice library — name, accent, gender, and use-case labels — so you can pick the right one for your project.
4. "Convert this FAQ into audio files, one per question." Claude splits the FAQ into individual questions and generates a separate MP3 for each, named sequentially (e.g., faq-01.mp3, faq-02.mp3).
5. "Generate a voice memo summary of these meeting notes." Claude first writes a condensed summary of your notes, then passes that summary to ElevenLabs to produce a listenable audio brief.

Pros

• No context switching — voice generation becomes a natural step in your Claude workflow instead of a separate app entirely.
• Free tier is sufficient for personal use — 10,000 characters/month covers blog summaries, short voiceovers, and daily audio memos without any cost.
• Officially maintained package — ElevenLabs maintains @elevenlabs/elevenlabs-mcp directly, so updates track new API features promptly.
• Pairs well with other servers — combine with a Zapier MCP server to automatically upload generated audio to cloud storage or email it to a client.

Cons / Limitations

• No real-time audio streaming — generated audio saves to a file rather than playing through your speakers immediately. You need to open the file in a media player.
• Character limits apply on free tier — 10,000 characters/month is roughly 70–80 minutes of audio. Heavy users will need a paid plan.
• Voice cloning requires a paid plan — the free tier gives access to the pre-made voice library only. Custom and cloned voices need Creator tier or above.