Free tool + setup
Clone your voice free, then give it to Claude Code
VoiceStudio is the free, open source ElevenLabs alternative. It clones your voice from a few seconds of audio, runs on your own computer, and ships with an MCP server so your agents can talk in your voice.
curl -fsSL https://voicestudio.sh/install | shWant our team to build it? See the done for you version
What it does
ElevenLabs, on your own machine
- Voice clone: give it 3 to 10 seconds of clean audio and it copies the voice.
- Speech, stories, audiobooks: type a script, it reads it back in that voice.
- 646 languages: same voice, different language.
- Voice design: set gender, age, pitch and accent, or describe a voice in plain English.
- MCP server: Claude Code, Codex and Cursor can clone, speak and transcribe with it.
Free and open source (AGPL-3.0). No account, and your audio stays on your computer. It only goes online to check for updates or download a model.
What you get
The files
- setup-prompt.txtPaste it into Claude Code. It checks your machine, installs VoiceStudio, connects the MCP server and makes a test clip in your voice.
- mcp.jsonThe Claude Code config, if you'd rather drop a file in your project (save it as .mcp.json).
- 00-README.mdThis whole page as one file.
Every file, with a download and a copy button, is in the full setup below.
The setup
Three steps, about 15 minutes
Step 1
Install VoiceStudio
One line in the Terminal. It downloads the official release, checks its checksum and puts the app in Applications.
macOS (Apple Silicon) or Linux curl -fsSL https://voicestudio.sh/install | shWindows (PowerShell) powershell -ep bypass -c "irm https://voicestudio.sh/install.ps1|iex"Or grab the installer from the releases page (opens in a new tab). You need about 8 GB of RAM and 10 GB of free disk. On a Mac it needs Apple Silicon and macOS 13.3 or newer.
Step 2
Clone your voice
Open VoiceStudio and click
Install local runtime(a few GB, one time). Install the default voice model when it asks (about 2.4 GB). Then openVoice cloning, add a clean recording of your voice, type a line and generate.Record somewhere quiet, one speaker, no music. 3 to 10 seconds is the sweet spot; longer doesn’t make it better. Only clone your own voice, or one you have permission to use.
Continue the setup
Unlock the full guide
Get every setup step, prompt, checklist, and local download.
By unlocking, you agree to get emails from QWave Labs about new free tools, tips and offers. Unsubscribe anytime. Privacy policy
Step 3
Hook its MCP server into Claude Code
Keep VoiceStudio open (the MCP server lives inside the running app) and run this once:
Terminal claude mcp add --transport http voicestudio http://localhost:3900/mcp/ --header "X-OmniVoice-Client-Id: claude-code"Check it with
claude mcp get voicestudio, or type/mcpinside Claude Code. Keep the trailing slash on/mcp/.Then just ask:
In Claude Code Clone my voice from ~/voice/me.wav, then read script.md in my voice and save it as voiceover.wav.
What Claude Code gets
The MCP tools
| Tool | What it does |
|---|---|
| clone_voice | A short recording in, a saved voice out. |
| generate_speech | Text in, audio in that voice out. |
| transcribe | Audio in, text out, 646 languages. |
| list_voices / list_languages | What's available on your machine. |
| check_health | Is VoiceStudio running, and on which chip. |
Want your agents writing, voicing and editing your content on their own? Our team sets that up. Book a call with our team
The lazy way
Let Claude Code set it all up
Paste this into Claude Code. It checks your machine, installs VoiceStudio, connects the MCP server and makes a test clip in your voice. It asks before any big download.
Install VoiceStudio on this computer and connect it to Claude Code.
1. Check the machine first: OS, chip, RAM, free disk. On a Mac it needs Apple Silicon,
macOS 13.3 or newer and about 10 GB free. Tell me if it doesn't fit and stop.
2. Install the desktop app with the official installer:
curl -fsSL https://voicestudio.sh/install | sh
(Windows: powershell -ep bypass -c "irm https://voicestudio.sh/install.ps1|iex")
3. Open VoiceStudio. On first launch I click "Install local runtime" myself. Wait until
http://localhost:3900/health answers {"status":"ok"} and tell me which device it uses.
4. Before downloading the voice model (k2-fsa/OmniVoice, about 2.4 GB), tell me the size
and ask me first.
5. Connect it to Claude Code:
claude mcp add --transport http voicestudio http://localhost:3900/mcp/ --header "X-OmniVoice-Client-Id: claude-code"
6. Check it: run claude mcp get voicestudio, then call the list_languages and check_health tools
and show me the result.
7. Ask me for a clean 3 to 10 second recording of MY voice, clone it with clone_voice,
generate one test sentence with generate_speech and save the WAV so I can listen.
Rules: only clone my own voice or a voice I have permission to use. Don't answer the
analytics prompt for me. Never print keys or tokens.
Free
The files
- setup-prompt.txtPaste it into Claude Code. It checks your machine, installs VoiceStudio, connects the MCP server and makes a test clip in your voice.Download
- mcp.jsonThe Claude Code config, if you'd rather drop a file in your project (save it as .mcp.json).Download
- 00-README.mdThis whole page as one file.Download
Before you start
The rules
- Only clone your own voice, or a voice you have permission to use.
- The open source build is AGPL-3.0. VoiceStudio sells a separate Pro licence if you need commercial terms.
- Intel Macs only get the app UI; the voice engine needs Apple Silicon, Windows or Linux.
Done for you
Want it done for you?
Our team builds the content machine around it: your voice, your scripts, your agents.
Book a call with our teamWho it’s for
Founders and creators who want voiceovers, clips and content made in their own voice without recording every take.
What you get
- Your voice cloned and set up on your machine
- Claude Code or Codex wired to write, voice and edit for you
- A review step so nothing goes out without your OK
- A walkthrough so you know exactly how to run it
What happens on the call
We look at how you make content today and tell you what we'd automate first. If it's a fit, we map out the build. If it's not, you still leave with the plan.