ElevenLabs MCP server adds speech, music, image, and video generation

• ElevenLabs has turned its MCP server into a multimodal generation surface for speech, transcription, dubbing, music, sound effects, images, and video, according to ElevenLabs' announcement.
• One installation brings ElevenLabs voice models, Scribe, dubbing, music, sound effects, and more than 50 image and video models into an assistant, ElevenLabs' client post says.
• The creative tools are available across Codex, Claude, Claude Code, ChatGPT, Cursor, GrokBot, Hermes, and OpenClaw in ElevenLabsDevs' availability post, while ElevenLabs' installation post points users to the add flow.
• Creative generation shares an installation with ElevenLabs Agents because ElevenLabs' Agents post says both run through the same MCP.

Scribe can turn an uploaded recording into a transcript for a script, captions, or a dub, as ElevenLabs' Scribe post describes. The hosted MCP setup page publishes the Claude Code endpoint, while the open-source MCP repository remains documented around speech and audio-processing tools.

The rollout exposes separate media operations that can also be composed into a single brief.

• Voiceover: Text to Speech can generate inline audio from any voice in the library, according to ElevenLabs' voiceover post.
• Transcription: Scribe accepts a recording and returns a transcript that can feed a script, captions, or a dub in ElevenLabs' transcription post.
• Dubbing: A request for another language runs through the same connector and returns a finished track, ElevenLabs' dubbing post says.
• Visual generation: The image and video path covers image generation, image edits, video animation, and lip-sync, per ElevenLabs' visual-tools post.
• Assembly: A single brief can return a script, voiceover, music bed, sound design, and video, as ElevenLabs' composition post demonstrates.

ElevenLabs has described the rollout with both a short client list and a broader developer-facing roster.

• ChatGPT, Claude, Cursor, Grok Bot, and Hermes are the clients in ElevenLabs' initial client list.
• Codex, Claude Code, Cursor, GrokBot, Hermes, OpenClaw, ChatGPT, and Claude appear as available in ElevenLabsDevs' post.
• Claude, ChatGPT, and Cursor are the three clients where more than 50 models are available in testingcatalog's overview.
• testingcatalog's follow-up calls Claude Code, ChatGPT, and Cursor connector targets, and describes Codex, OpenClaw, and Hermes as companion skills.

The official MCP page calls the service a hosted MCP server. Its Claude integration uses one-time, revocable OAuth rather than an API key, and its Claude Code registration command is:

The page says /mcp inside a Claude Code session completes sign-in. ElevenLabs' post on the shared MCP adds that the same connector powers ElevenLabs Agents.

The original ElevenLabs MCP launch post framed the project around local tools for voice agents, transcription, and speech output. The repository README still describes its public server in terms of text-to-speech and audio processing, while the hosted rollout supplies the newer client-side creative entry points.

Generated assets are stored in an ElevenCreative workspace, ElevenLabs' workspace post says. Studio can then open them for timeline edits, narration refinement, music and sound-effect layering, and export.