ElevenLabs MCP: Documentation
Updated:
Connection
uvx elevenlabs-mcp (recommended) or pip install elevenlabs-mcp (stdio). The ELEVENLABS_API_KEY env var holds a key from your ElevenLabs account. Ready-made configs exist for Claude Desktop, Cursor, Windsurf and OpenAI Agents. Extra env vars set the output mode (disk file / MCP resource / both), the data-residency region and a base path for files.
Tools
- Text-to-speech — generate audio from text using a library voice or a cloned one.
- Voice cloning — build a voice model from a sample recording.
- Voice design — generate a brand-new voice from a text description of tone and character.
- Transcription (speech-to-text) — recognize speech from an audio file.
- Audio isolation — clean a voice track of background noise and music.
- Voice variation generation to help pick the right tone.
Agent flow
"Read this text aloud in Rachel's voice and save it as mp3", "transcribe this call recording", "design a new voice — calm male tone with a slight accent", "remove the background noise from this track".
Pricing
Free — 10,000 credits a month. Paid ElevenLabs plans lift the credit cap and add commercial rights to generated voices; the MCP server itself is open source and free (1.5k★).