Summary

Scripts, not just single lines — and a history that outlives the session.

The multi-part script job is what separates it from a one-shot speak call: a dialogue with a different voice per part is one submission, and the SQLite history means yesterday's renders are still findable.

What it is

A text-to-speech server for the ElevenLabs API with a persistent history. Single lines and full scripts are both supported, generated files are retrievable afterwards, and every job is kept in a local SQLite database.

What you get

A line of text spoken in your configured voice, in one call
A multi-part script generated as a job, with parts able to use different voices
The voices available on your account listed, so a voice is chosen rather than guessed
Generated audio files fetched back by reference
A voiceover history that survives restarts, and jobs deleted from it when they are no longer wanted

Requirements

Uvx elevenlabs-mcp-server, with no separate install needed. An ElevenLabs API key as ELEVENLABS_API_KEY, plus ELEVENLABS_VOICE_ID and ELEVENLABS_MODEL_ID; stability, similarity boost and style sit in the same config, and ELEVENLABS_OUTPUT_DIR decides where files land. A sample SvelteKit client ships with the repo if you want a browser view of the history.

Setup effort

One command plus a key — npx -y @smithery/cli install elevenlabs-mcp-server --client claude, then supply credentials