ElevenLabs MCP Server: Tools, Setup and Security

The ElevenLabs MCP server lets an AI app create and manage ElevenLabs voice agents and generate speech. It can update agents, manage their knowledge base, run tests and work with deployments. ElevenLabs hosts it at https://api.elevenlabs.io/v1/mcp, and it signs in with OAuth.

Checked against: ElevenLabs hosted MCP server docs on 28 October 2026. eSEOspace did not install or test this server.

At a glance

What it can do

The docs describe capabilities rather than tool names. They include:

• Creating, updating, listing, duplicating and deleting agents.
• Managing procedures, tools, MCP servers and knowledge base documents.
• Running tests.
• Working with branches, drafts and deployments.
• Reading and updating triage tickets.
• Retrieving widget configuration.
• Generating speech audio.

OAuth scopes cover ElevenAgents read and write operations and Text to Speech. Because tool names are not published in the docs, check your client's tool list after connecting.

How to connect it

For Claude Desktop, the docs give three steps:

1. Go to Settings > Connectors and browse the directory.
2. Search for ElevenLabs and select Connect.
3. Complete the OAuth sign-in.

For data residency regions, add a custom connector with your region's URL instead:

• EU: https://api.eu.residency.elevenlabs.io/v1/mcp
• India: https://api.in.residency.elevenlabs.io/v1/mcp
• Singapore: https://api.sg.residency.elevenlabs.io/v1/mcp

For other MCP clients, the docs say to point the client at your region's URL and complete OAuth. Clients that support hosted client metadata need no separate registration. Our page on client ID metadata documents explains how that works.

The old local server on GitHub now opens with a warning. It says the local server "is deprecated in favor of the ElevenLabs hosted MCP server" and the repository "is no longer actively maintained."

Security notes

Changes go live. The docs warn: "Updating an agent through the MCP server publishes the change directly." Changes land on the target branch or main by default and are not staged as drafts. If a live voice agent talks to your customers, test changes on a branch first.

Deletes. The docs say "Deleting an agent is destructive." They advise reviewing tool calls before approving them.

Per-tool controls. In your MCP client you can turn individual tools on or off and choose whether each runs automatically or asks first. Workspace admins can set this for the organization. "A tool disabled by an administrator cannot be re-enabled by an individual user."

Scoped access. Access is limited to what you approve on the OAuth consent screen, and you can revoke it any time. The docs also advise restricting write access to members who need it. See excessive permissions in MCP.

For broader guidance, visit the MCP security center.

Pricing

The MCP docs make no billing statements. ElevenLabs' pricing page lists a Free plan with 10k credits a month and paid plans from Starter to Enterprise, as of October 2026. Speech generation and agent usage draw on your plan.

Alternatives

• Hugging Face MCP server: can call community Gradio Spaces, including audio models.
• Perplexity MCP server: if your agent needs web-grounded answers rather than voice.

More on our MCP server profiles hub.

Next step

A voice agent that edits itself in production needs clear rules. Our Claude Code consulting and MCP readiness audit help teams set approvals and review steps before they connect live systems.

Frequently asked questions