One sign-in puts 50+ ElevenLabs models inside Claude

No API key to paste, nothing to keep running on your machine: ElevenLabs has connected its hosted MCP server to ElevenCreative, so an assistant can make speech, music, sound effects, images and video inside the conversation where you asked for them. Testingcatalog reports that more than 50 models now sit behind one sign-in, reachable from Claude, ChatGPT and Cursor through the connector ElevenLabs already ships.

At a glance

• The connector that already shipped for Claude, ChatGPT and Cursor now carries the creative tools in the same list, so drafting and producing an asset happen in one session instead of two applications.
• Installation runs on OAuth: Claude installs it from the connectors directory, Claude Code with one terminal command, ChatGPT as a developer mode plugin, Cursor from settings, scoped to one workspace.
• The announcement says nothing about what generation through the connector costs, and access permissions range from read-only monitoring through to full agent management, revocable from the client or from ElevenLabs account settings.

If you have not been following: ElevenLabs was founded in 2022 by Piotr Dabkowski and Mati Staniszewski, built its name on speech synthesis and dubbing, and later split the range into three lines. ElevenCreative covers generation, ElevenAgents covers voice and chat agents, ElevenAPI covers direct programmatic access. The hosted connector arrived in August, and it covered agents only.

The August connector could clone an agent but not say a word

Since August the hosted connector did management work. It could surface the topics coming up in recent conversations, duplicate an agent for a new market, compare configurations across a workspace, and estimate expected model usage and cost before a change shipped. Generation stayed in the ElevenLabs web app or behind the API.

ElevenCreative is the brand ElevenLabs put on its generation products when it grouped them earlier this year: text-to-speech, speech-to-text, voice changer, voice cloning, voice isolation, sound effects, music, voice design, the Studio editor for long-form projects, the Ads Engine, and the image and video generators. All of it now sits in the same tools list as the agent commands, so a script, a voiceover, a soundtrack and a cover image can come out of one session.

Installation runs on OAuth instead of API keys

In Claude the connector installs from the connectors directory in a few clicks. Claude Code adds it from the terminal with a single command, ChatGPT takes it as a developer mode plugin, and Cursor accepts the server configuration in settings.

Access is scoped to one ElevenLabs workspace and can be revoked from either the client or ElevenLabs account settings. Permissions run from read-only monitoring through to full agent management, so an admin decides how much of a workspace a chat session can touch.

There is a second path for coding agents: a command-line install through npm, shipped alongside a set of companion skills, covering Codex, Claude Code, OpenClaw and Hermes. The original ElevenLabs MCP server from 2025, the local one that runs on a developer machine with a user-supplied API key, stays available for developer workflows.

Why does any of this need a protocol?

MCP is short for Model Context Protocol, an open standard that lets an assistant call outside services on a user's behalf. The client, meaning your chat app or your IDE, talks to a server that exposes a set of capabilities; according to Codingscape, those servers are lightweight programs that can reach local data or remote services over the internet.

Anthropic, which open-sourced the specification, describes it as a universal open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol, and compares it to a USB-C port. ElevenLabs hosts this server itself, so the sign-in is the connection, and there is no key sitting in a config file to rotate or leak.

Oddly, a release that puts more than 50 models one prompt away says nothing about what a generation through the connector costs or how it meters against a workspace plan, and that is the number most teams would check first. The permissions range is broad as well: the inputs describe everything from read-only monitoring to full agent management, but do not say which level the creative tools need before an assistant can render video on your account.

Whether drafting becomes the studio

Testingcatalog reads the addition as a bet that production shifts toward the writing surface, where an audio or video asset gets made during drafting instead of commissioned once the script is finished. That bet will be tested on invoices, and no price has been published for generation through the connector. The other open question is whether the hosted server and the 2025 local one stay two separate paths for long.