# Create voiceovers for your videos

> Recording voiceovers for every edit and language takes a studio and an afternoon.

This setup is for video editors, course creators and marketers who need voiceovers for drafts, revisions and other languages. Claude reads your script from the project folder, sends it to ElevenLabs text to speech with the voice and settings you choose, and saves the audio file next to your edit. You end up with numbered audio files you can drop into your timeline and regenerate whenever the script changes.

## Steps

1. **Connect ElevenLabs**: Create an ElevenLabs API key in your account settings and add the official ElevenLabs MCP server to Claude Desktop or Claude Code, usually run with uvx elevenlabs-mcp and the key in ELEVENLABS_API_KEY. Set ELEVENLABS_MCP_BASE_PATH to your project's audio folder so files save in the right place. Ask Claude to list a few available voices to confirm the connection.
   Tip: Each generation uses ElevenLabs credits from your plan, so test with short lines first.
2. **Give it the script**: Add the Filesystem MCP server with access to your project folder only, for example the folder that holds the script and the edit. Ask Claude to read the script file and split it into numbered lines or scenes. Check the split before generating, since each chunk becomes one audio file.
   Tip: Keep scripts in plain text or markdown so Claude can read them without conversion.
3. **Generate and save**: Tell Claude which voice, language and pacing to use, then ask it to generate each chunk and save it with a clear name such as 03_intro_en.mp3. Listen to the files and ask Claude to regenerate only the lines that sound wrong. For another language, give Claude a reviewed translation of the script, or ask it to translate and then check the text before generating.
   Tip: Note the voice ID and settings you liked in a text file in the project so later edits match.

## Prompts to try

- Read script.md in the project folder, split it by scene and generate each scene with a calm, mid-paced British male voice. Save as numbered MP3 files in /audio.
- Regenerate line 7 more slowly and with less emphasis on the product name.
- Translate the script into Spanish, show me the text first, then generate it with the same voice.
- List voices that suit a friendly product tutorial and show me their IDs.
- The script changed in scenes 2 and 5. Regenerate only those two files.

## What you need

- ElevenLabs account and API key
- Claude Desktop or Claude Code
- uv (for uvx) and Node.js installed locally
- Script saved as a text file in the project folder

## Limits

- Pronunciation of names, acronyms and jargon may need respelling or several tries.
- Claude cannot hear the result; you judge the timing and tone yourself.
- Generated audio is not automatically synced to your video timeline.

## Security notes

- Only clone voices you have permission to use.
- Limit file access to the project folder.

## MCP servers

- [ElevenLabs MCP Server](https://github.com/elevenlabs/elevenlabs-mcp): generate-audio, voice-read
- [Filesystem MCP Server](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem): file-read, file-write

## FAQ

**What does it cost?**

ElevenLabs charges by credits based on the number of characters you generate, and your plan sets the monthly allowance. Regenerating a whole long script many times adds up, so regenerate only the lines that change.

**Can I clone my own voice?**

ElevenLabs supports voice cloning, depending on your plan, and the MCP server exposes cloning tools. Only clone voices you own or have written permission to use.

**Can I use the audio commercially?**

Commercial use depends on your ElevenLabs plan and its terms. Check the current terms before using generated audio in paid work.

**Why limit the Filesystem server to one folder?**

The Filesystem server can read and write any directory you allow. Restricting it to the project folder keeps Claude away from your other files.

## Tools

- **[ElevenLabs](https://suiteintent.com/t/elevenlabs?m=GB)**: AI text to speech with lifelike voices. · Best for: Narration and dubbing in many languages · Free plan · 17%