ElevenLabs Agents MCP Server

Official ElevenLabs MCP server for building conversational AI voice agents, generating speech, cloning voices, transcribing audio, and placing outbound calls.

AI/ML by ElevenLabs API Key active

GitHub Docs

Overview

The ElevenLabs MCP server is the official integration that exposes ElevenLabs' text to speech, speech to text, voice cloning, and Conversational AI Agents platform to MCP clients like Claude Desktop, Cursor, Windsurf, and OpenAI Agents. It acts as a local abstraction over the ElevenLabs API, letting an LLM create voice agents, attach knowledge bases, place outbound phone calls, transcribe audio, generate sound effects, and synthesize speech without writing API client code.

The Agents capabilities are first-class: you can create agents with custom prompts and voices, list and inspect existing agents, attach knowledge bases from URLs, files, or raw text, list and read conversation transcripts (with analysis), and trigger outbound calls through linked phone numbers. Alongside the agent tools, the server provides the full ElevenLabs audio toolbox: text_to_speech, speech_to_text with diarization, voice cloning, voice isolation, speech to speech conversion, sound effect generation, and music composition.

Authentication is a simple API key (free tier ships with 10k credits per month). The server is published on PyPI as elevenlabs-mcp and is typically launched via uvx. Because most operations write audio files, the server supports configurable output paths and resource modes.

Tools

Tool	Description
`create_agent`	Create a conversational AI agent with a custom system prompt, voice, and model configuration.
`list_agents`	List all conversational AI agents on the account.
`get_agent`	Retrieve full configuration and metadata for a specific agent.
`add_knowledge_base_to_agent`	Attach a knowledge base to an agent via URL, file, or raw text.
`list_conversations`	List agent conversations with filtering and pagination.
`get_conversation`	Fetch a conversation's full transcript and analysis.
`make_outbound_call`	Trigger an outbound phone call from an agent using a linked phone number.
`list_phone_numbers`	List phone numbers linked to the account for outbound calling.
`text_to_speech`	Convert text to speech with a chosen voice, model, and audio parameters.
`speech_to_text`	Transcribe audio files, optionally with speaker diarization.
`text_to_sound_effects`	Generate short sound effects (0.5 to 5 seconds) from a text description.
`text_to_voice`	Generate three preview voice variations from a text description.
`create_voice_from_preview`	Save one of the previewed voice variations as a usable voice.
`voice_clone`	Create an instant voice clone from one or more audio samples.
`search_voices`	Search the user's voices by name, description, labels, or category.
`search_voice_library`	Search the public ElevenLabs voice library.
`add_voice_to_library`	Add a shared voice from the public library to the user's voices.
`get_voice`	Retrieve details for a specific voice.
`list_models`	List all available speech synthesis models.
`speech_to_speech`	Transform an audio file to a different voice while keeping the original delivery.
`isolate_audio`	Isolate voice from background noise in a mixed audio file.
`play_audio`	Play an audio file from the local filesystem.
`compose_music`	Generate music from a prompt.
`create_composition_plan`	Create a structured composition plan for music generation.
`check_subscription`	Return current subscription status and API usage metrics.

Setup Guide

Prerequisites

An ElevenLabs account and API key (free tier includes 10k credits per month, available from the ElevenLabs dashboard).
uv installed: curl -LsSf https://astral.sh/uv/install.sh | sh
On Windows with Claude Desktop, enable Developer Mode.

Install

The server is published to PyPI and run with uvx:

pip install elevenlabs-mcp

Or run directly without installing:

uvx elevenlabs-mcp

MCP client config

Add the following to claude_desktop_config.json (or your client's MCP config file):

{
  "mcpServers": {
    "ElevenLabs": {
      "command": "uvx",
      "args": ["elevenlabs-mcp"],
      "env": {
        "ELEVENLABS_API_KEY": "<insert-your-api-key-here>"
      }
    }
  }
}

Optional environment variables

ELEVENLABS_MCP_BASE_PATH: default directory for generated audio files (defaults to ~/Desktop).
ELEVENLABS_MCP_OUTPUT_MODE: files, resources, or both (defaults to files).
ELEVENLABS_API_RESIDENCY: data residency region, enterprise only (defaults to us).

Use Cases

Spin up a customer support voice agent with a custom system prompt, attach product docs as a knowledge base, then list and review conversation transcripts.
Place outbound calls from a sales agent using a linked phone number and review the resulting transcript and analysis afterwards.
Generate narration for a video by passing the script to text_to_speech with a cloned voice and writing the file to disk.
Transcribe meeting recordings with speaker diarization, then summarize speakers' contributions inside the same chat.
Prototype a brand voice by previewing variations with text_to_voice, picking one, and saving it for reuse across agents and TTS calls.

Example Prompts

"Create a new ElevenLabs agent named 'Support Bot' that answers in a friendly tone, then attach https://docs.example.com as its knowledge base."
"List my last 20 conversations for agent_id abc123 and summarize the top reasons customers called."
"Make an outbound call from my support agent to +14155551212 and pull the transcript when it's done."
"Transcribe ./meeting.mp3 with speaker diarization and give me action items per speaker."
"Clone my voice from these three samples and then read this blog post aloud using the new voice."

Pros

Official server maintained by ElevenLabs, kept in sync with the platform API.
Broad surface area: Agents, TTS, STT, voice cloning, voice library, sound effects, music, and outbound calls in one server.
Simple API key auth and a generous free tier (10k credits/month) for evaluation.
Configurable output modes (files, resources, both) make it easy to integrate with different client UIs.

Limitations

All operations consume ElevenLabs credits, so heavy use requires a paid plan.
Some operations (voice design, audio isolation, outbound calls) can be long-running and may trip MCP client timeouts even when the backend succeeds.
Data residency (non-US) is enterprise only, and Windows users must enable Developer Mode in Claude Desktop.

Alternatives

jezweb/elevenlabs-mcp-server: community MCP server focused specifically on Conversational AI agent and knowledge base management.
199-mcp/mcp-elevenlabs: community fork of the official server with additional conversation history features.
mamertofabian/elevenlabs-mcp-server: earlier community implementation centered on TTS workflows.