This Free Tool Just Killed ElevenLabs (And Made Claude Talk Back To You)

TL;DR: Voicebox is a free, open-source app that clones your voice from a 10-second recording and speaks it back in 23 languages, entirely on your own computer. Connect it to Claude Code with one MCP command and your AI agent starts talking back to you out loud, in your own voice, when it finishes a task. It also replaces your paid dictation app. No subscription, no cloud upload, no catch beyond a Gatekeeper warning you clear in one terminal command.

The problem with paying for a voice

If you have ever cloned your voice with ElevenLabs, you know the drill: pick a plan, upload samples, wait for processing, watch the character-credit meter tick down every time you generate something. It works, but it is a subscription for something your own laptop can now do for free.

Separately, if you use a dictation app like Wispr Flow to talk instead of type, that is a second subscription, for a second job your computer can also do on its own.

Voicebox, a free open-source app built by Jamie Pine (the developer behind Spacedrive), does both jobs at once, for free, and it never sends your voice anywhere. It has quietly picked up 39,922 GitHub stars in about five and a half months, which is the kind of growth number that usually means a lot of people found the same shortcut around the same two subscriptions.

What Voicebox actually does

Three things, and they stack:

1. Clones your voice locally. Record about 10 seconds of yourself talking, and Voicebox creates a voice profile you can use to generate speech in your own voice from any text you type.
2. Speaks that voice in 23 languages. One specific engine inside Voicebox, called Chatterbox Multilingual, is the one doing this. It is the only engine in the app that can both clone a voice from a sample AND speak it in all 23 supported languages. (Another engine, Qwen3-TTS, can clone a voice too, but it tops out at 10 languages. Worth knowing if you go exploring the other engines yourself.)
3. Gives your AI a voice. Voicebox ships an MCP server, which means you can connect it directly to Claude Code. Once connected, Claude can call a tool named voicebox.speak to talk out loud in your cloned voice, the second it finishes a task.

That third point is the one ElevenLabs cannot touch, because ElevenLabs is a voice generation API, not something that plugs into your coding agent’s workflow out of the box the way this does.

Setting it up

The honest version: this takes about 10 minutes, most of which is a model downloading in the background.

Install it. Voicebox pulls straight from the latest GitHub release for Apple Silicon.

curl -L -o ~/Downloads/Voicebox.app.tar.gz https://github.com/jamiepine/voicebox/releases/latest/download/Voicebox_aarch64.app.tar.gz
tar -xzf ~/Downloads/Voicebox.app.tar.gz -C ~/Downloads/
mv ~/Downloads/Voicebox.app /Applications/

Clear the Gatekeeper warning. Voicebox is not notarized with an Apple Developer certificate yet, so macOS will call it “damaged.” It is not. This one command strips the quarantine flag:

xattr -cr /Applications/Voicebox.app
open /Applications/Voicebox.app

Create your voice profile. Inside the app: Profiles, then + New Profile, name it, set the engine to Chatterbox Multilingual, then Record Sample. Speak naturally for 10 to 30 seconds, clean audio, no background noise. Anything under 5 seconds gives you a noticeably worse clone, so do not rush this part. Click Create Profile and it saves instantly.

Type something and hear yourself say it. Go to the Generate tab, paste in a sentence, hit Generate, then Play. Switch the language dropdown to Spanish, or any of the other 22 languages, and generate the same voice again. Same clone, new language.

Connecting it to Claude so it talks back

This is the part that turns a voice cloning app into something that changes how you work. Voicebox must already be running for this to work, since its local server only listens while the desktop app is open.

Run this one-liner in your terminal:

claude mcp add voicebox \
  --transport http \
  --url http://127.0.0.1:17493/mcp \
  --header "X-Voicebox-Client-Id: claude-code"

Or drop this straight into a project’s .mcp.json:

{
  "mcpServers": {
    "voicebox": {
      "type": "http",
      "url": "http://127.0.0.1:17493/mcp",
      "headers": {
        "X-Voicebox-Client-Id": "claude-code"
      }
    }
  }
}

Then add one line to your CLAUDE.md so it happens automatically on every task:

After completing any task, call the voicebox.speak MCP tool with a one-sentence summary of what you did, so I hear it read back to me.

Now when Claude finishes a task, it does not just print text and wait. It says “done” out loud, in your own cloned voice.

Want the full copy-paste setup guide, including the exact xattr fix, model pre-cache commands, and MCP wiring? Grab the free guide here.

The second app it quietly kills

Voicebox is not just a voice cloning tool. It also ships free, offline dictation for your entire computer, no config required. Hold Right Command plus Right Option anywhere on your Mac, talk, release, and the transcript pastes into whatever field had focus. Tap Space while holding to switch to hands-free toggle mode.

That is a direct replacement for a dictation subscription, running on the same free app that is already cloning your voice.

The honest caveats

A few things worth knowing before you rely on this for anything important:

• First generation on an uncached model takes 2 to 5 minutes. Pre-cache the model before you need it, or you will be staring at a spinner.
• Voicebox has to already be running for the Claude MCP connection to work. If the desktop app is not open, the voicebox.speak call just fails silently.
• No Linux build yet. macOS and Windows only, as of this writing.
• “Clone any voice into 23 languages” is specifically true of the Chatterbox Multilingual engine, not every engine in the app. Keep that distinction in mind if you explore the other TTS options Voicebox ships with.

None of these are dealbreakers. They are just the things that will trip up a first-time setup if nobody tells you in advance.

Why this probably killed ElevenLabs for a lot of people

ElevenLabs solved a real problem: cloning a voice used to require real engineering. It built a great product and a business around that gap. But the gap only stays a business while nobody ships the same capability for free, running locally, with no subscription. Voicebox does that, plus it plugs directly into the coding agent a lot of people are already running every day.

That does not mean ElevenLabs is going away. Its API, its production tooling, and its polish still matter for teams shipping voice products at scale. But for the person who just wants their own voice cloned, in their own languages, without paying monthly for it, and who wants their AI agent to actually talk back to them, Voicebox is very hard to argue with.

FAQ

Is Voicebox actually free? Yes. It is open-source and runs entirely on your own computer. There is no subscription, no character limit, and no cloud processing fee.

Does my voice data ever leave my computer? No. Voicebox runs 100% locally. Nothing you record or generate is uploaded anywhere.

Can Voicebox really speak 23 languages? Yes, specifically through its Chatterbox Multilingual engine, which is the only engine in the app that both clones a voice and supports all 23 languages.

Do I need to know how to code to set this up? No. The install is three terminal commands you copy and paste, and the rest happens inside the app. The Claude Code connection is one more command if you want that part.

Does this replace my dictation app too? Yes. Voicebox ships a system-wide dictation hotkey (Right Command plus Right Option) that works offline, in any app, for free.

---

Save this and send it to anyone still paying for ElevenLabs or a dictation app. If you want the full copy-paste setup, comment CLONE on the video and it’ll land in your DMs, or join the Actionable AI community for a new setup like this every week.