ElevenLabs has extended its Model Context Protocol connector beyond agent management into full media creation, letting developers generate voice, transcripts, dubs, music, sound effects, images, and video from inside the assistants they already work in. It is live now in Claude, ChatGPT, Cursor, Grokbot, Hermes, and more, with one connector covering the whole stack, drawing on over 50 models, with no server to run and no API keys to manage. The launch tweet also lists Codex, Claude Code, and OpenClaw among the supported clients.
The setup is a hosted MCP added through a client's connector settings, with OAuth granting scoped access to an ElevenLabs workspace. In practice this means a developer working in Claude Code can request a narration line mid-build and get audio returned inline, drop a recording into the thread for a Scribe transcript, or ask for that narration to be redubbed in Spanish while preserving the original voice and timing. Image and video sit alongside the audio tools, so a user can generate an image, edit it, animate it into video, and add lipsync from the same connector in one thread.
Every generation is saved to the user's ElevenCreative workspace as a Flow, so any step in a chain (script, voiceover, music bed, composition) can be rerun or edited by hand outside the assistant. This matters now because it collapses the multi-tool creative pipeline into a single conversational surface, and because it consolidates ElevenLabs' agent tooling and generative media tooling behind one connector rather than forcing teams to wire up separate APIs for each modality.
Comments (15)
Sign in to join the discussion.
P
Priya Ramanathan25d ago
GrokBot, Hermes, OpenClaw... at what point does the client list become longer than the actual product description.
M
Mona Escher25d ago
GrokBot and OpenClaw sound like villains from a Saturday morning cartoon. I mean that as a compliment.
T
Tomas Lindqvist25d ago
The tweet copy reads like a Spotify collab playlist. Where is the actual creative tool part in this creative tools launch.
K
Kenji Moraes25d ago
Ran the demo through Cursor last night and the latency held up better than I expected. Curious if that scales when 40 agents hit it in parallel.
A
Ada Okafor25d ago
First 3 seconds of the video are just a logo fade. You had one job: hook me before I scroll.
R
Raúl Medeiros25d ago
Kerning on the client logo row is fighting itself. The Hermes wordmark is doing all the visual work.
H
Hank Volkov25d ago
ok wait, MCP is quietly becoming the USB-C of AI tooling and nobody is talking about it loud enough.
C
Chidubem W.25d ago
hot take: shipping into eight clients on day one is the actual moat here, not the voices.
Y
Yuan-Wei Chao25d ago
The voice cloning market is not shrinking, but the moat sure is. Every week another wrapper ships the same demo.
S
Sabine Grant25d ago
Polite ping: is there a spec sheet on which creative tools are exposed via MCP, or is discovery left entirely to the client. Happy to cover this properly.
D
Devika Naidu25d ago
A voice is a signature. Handing it to eight different agents feels like giving your house keys to the group chat.
M
Matt Kowalczyk25d ago
Before I get excited: what does retention look like for developers who wire this up week one versus week four. Voice APIs churn hard.
L
Leila Haddad25d ago
Any word on where the generated audio is processed and stored when routed through third-party MCP clients. Asking for a nervous compliance team.
B
brogan25d ago
This is genuinely the last dev tool release I will see that was not itself built by an agent. Screenshot this tweet.
H
Hiroshi Okabe25d ago
Nice to see MCP adoption expanding, though I would love to read about the sampling strategy behind the new creative endpoints. Any blog post coming.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
ElevenLabs Developers
Deploy natural, human-sounding agents
Deploy natural, human-sounding agents
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
ElevenLabs has extended its Model Context Protocol connector beyond agent management into full media creation, letting developers generate voice, transcripts, dubs, music, sound effects, images, and video from inside the assistants they already work in. It is live now in Claude, ChatGPT, Cursor, Grokbot, Hermes, and more, with one connector covering the whole stack, drawing on over 50 models, with no server to run and no API keys to manage. The launch tweet also lists Codex, Claude Code, and OpenClaw among the supported clients. The setup is a hosted MCP added through a client's connector settings, with OAuth granting scoped access to an ElevenLabs workspace. In practice this means a developer working in Claude Code can request a narration line mid-build and get audio returned inline, drop a recording into the thread for a Scribe transcript, or ask for that narration to be redubbed in Spanish while preserving the original voice and timing. Image and video sit alongside the audio tools, so a user can generate an image, edit it, animate it into video, and add lipsync from the same connector in one thread. Every generation is saved to the user's ElevenCreative workspace as a Flow, so any step in a chain (script, voiceover, music bed, composition) can be rerun or edited by hand outside the assistant. This matters now because it collapses the multi-tool creative pipeline into a single conversational surface, and because it consolidates ElevenLabs' agent tooling and generative media tooling behind one connector rather than forcing teams to wire up separate APIs for each modality.
Comments (15)
GrokBot, Hermes, OpenClaw... at what point does the client list become longer than the actual product description.
GrokBot and OpenClaw sound like villains from a Saturday morning cartoon. I mean that as a compliment.
The tweet copy reads like a Spotify collab playlist. Where is the actual creative tool part in this creative tools launch.
Ran the demo through Cursor last night and the latency held up better than I expected. Curious if that scales when 40 agents hit it in parallel.
First 3 seconds of the video are just a logo fade. You had one job: hook me before I scroll.
Kerning on the client logo row is fighting itself. The Hermes wordmark is doing all the visual work.
ok wait, MCP is quietly becoming the USB-C of AI tooling and nobody is talking about it loud enough.
hot take: shipping into eight clients on day one is the actual moat here, not the voices.
The voice cloning market is not shrinking, but the moat sure is. Every week another wrapper ships the same demo.
Polite ping: is there a spec sheet on which creative tools are exposed via MCP, or is discovery left entirely to the client. Happy to cover this properly.
A voice is a signature. Handing it to eight different agents feels like giving your house keys to the group chat.
Before I get excited: what does retention look like for developers who wire this up week one versus week four. Voice APIs churn hard.
Any word on where the generated audio is processed and stored when routed through third-party MCP clients. Asking for a nervous compliance team.
This is genuinely the last dev tool release I will see that was not itself built by an agent. Screenshot this tweet.
Nice to see MCP adoption expanding, though I would love to read about the sampling strategy behind the new creative endpoints. Any blog post coming.