ElevenLabs MCP now powers Claude

Jeff Liu··2 min read·AI
ElevenLabs MCP now powers Claude
ListenElevenLabs MCP now powers Claude
0:00
--:--

Key Takeaways

  1. 1ElevenLabs launched its hosted MCP connector on August 17, 2026, integrating AI agents with Claude.
  2. 2Claude now manages ElevenLabs voice agents directly using natural language commands.
  3. 3Claude can estimate LLM usage and costs for agent changes before deployment.
  4. 4The new MCP uses OAuth for secure authentication, replacing local API keys.

ElevenLabs introduced a hosted Model Context Protocol (MCP) connector on August 17, 2026, enabling its AI, Claude, to access and manage voice and chat agents built with ElevenLabs Agent technology directly from Claude.

This integration streamlines the management and optimization of AI agents within the Claude environment. Developers can now inspect, modify, and monitor their production voice agents using natural language commands.

The update focuses on enhancing workflow efficiency and providing greater control over agent configurations.

What Does the MCP Connector Enable?

The ElevenLabs MCP connector grants Claude read and write access to ElevenAgents, allowing for comprehensive management tasks such as configuration, monitoring, and performance improvement directly within the AI assistant. Claude can also estimate Large Language Model (LLM) usage and cost before implementing changes.

This means Claude can perform various actions. Users can ask it to surface topics from recent conversations for analysis or to duplicate an agent for a new market by updating its voice and initial message.

The connector allows Claude to compare configurations across agents and flag inconsistencies. It also provides cost estimations, like determining how a checkout agent's cost per conversation would change when switching from GPT-4o to Gemini 2.5 Flash.

How Does Authentication Evolve?

The new hosted MCP connector utilizes OAuth for secure authentication, moving away from the previous requirement for local API keys and simplifying secure access to agent management. This change enhances security and ease of use by centralizing control over connections.

ElevenLabs initially released an open-source MCP server in April 2025. This original server required developers to run it locally and use their own API keys for functions like generating speech or cloning voices for Claude Desktop or Cursor.

The new hosted MCP, however, manages agents already within an ElevenLabs workspace. Users sign in securely using OAuth, eliminating API key management and allowing connections to be revoked anytime. This approach provides a managed and secure solution for agent operations.

Feature

Original MCP Server (2025)

New Hosted MCP (2026)

Deployment

Local, developer-managed

Hosted by ElevenLabs

Authentication

API keys

OAuth

Focus

Developer workflows (speech, cloning)

Managing existing ElevenAgents

Control

Local API key management

Secure sign-in, revoke via account settings

Related Articles

More insights on trending topics and technology

The Signal

What shipped in AI this week, with the sources.

One email a week.