You can now build and deploy AI voice agents that speak, listen, and understand — all from your Claude chat. ElevenLabs MCP server brings the full power of voice synthesis, cloning, transcription, and outbound calling directly into your AI client.
The MCP server acts as a bridge between your AI assistant and ElevenLabs' audio platform, allowing you to create conversational AI agents, perform outbound calls, transcribe speech, and generate audio — all with simple prompts.
This guide walks you through building your first AI voice agent with ElevenLabs MCP and Claude, step by step. Get your free ElevenLabs API key here.
⚡ TL;DR — Build Voice Agents from Your Chat
Step 1: Get your ElevenLabs API key (free tier with 10,000 credits). Step 2: Install the MCP server in Claude Desktop with uv. Step 3: Ask Claude to "Create an AI agent that speaks like a film noir detective." Step 4: Configure voice, personality, language, and behavior. Step 5: Test your agent and deploy it for outbound calls. Start your free trial here.
What Is ElevenLabs MCP?
ElevenLabs MCP is a Model Context Protocol server that connects your AI agent to ElevenLabs' full audio platform. It works with Claude Desktop, Cursor, Windsurf, OpenAI Agents, and other MCP clients.
The server enables you to:
- Generate speech — Create high-quality voiceovers in thousands of voices
- Clone voices — Convert recordings to sound like different characters or speakers
- Transcribe audio — Turn speech into text and identify different speakers
- Create sound effects — Generate soundscapes from text descriptions
- Manage agents — Build, configure, and deploy voice agents
- Make outbound calls — Order pizza, book appointments, or follow up with leads
Step-by-Step: Build Your First AI Voice Agent
Step 1: Get Your ElevenLabs API Key
Sign up for ElevenLabs and get your API key from the dashboard. The free tier gives you 10,000 credits to start. Get your free API key here.
Step 2: Install the MCP Server
First, install uv (Python package manager):
curl -LsSf https://astral.sh/uv/install.sh | sh
Then, configure Claude Desktop by adding the following to your claude_desktop_config.json file:
{
"mcpServers": {
"ElevenLabs": {
"command": "uvx",
"args": ["elevenlabs-mcp"],
"env": {
"ELEVENLABS_API_KEY": "your-api-key-here"
}
}
}
}
On macOS, the config file is at ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows, it's at %APPDATA%Claudeclaude_desktop_config.json. If you're using Windows, enable Developer Mode in Claude Desktop from Help > Enable Developer Mode.
Step 3: Ask Claude to Create Your Agent
Once the MCP server is connected, simply ask Claude in natural language:
"Create an AI agent that speaks like a film noir detective and can answer questions about classic movies."
Claude will open an interactive UI where you can:
- Set the agent name — Give your agent a memorable name
- Write a system prompt — Define the agent's personality and behavior
- Choose a voice — Browse and preview ElevenLabs' voice library
- Select language — Choose from 32 supported languages
- Adjust voice settings — Control stability, similarity boost, and temperature
Step 4: Configure Your Agent
The agent creator form gives you full control over your voice agent:
- Voice Configuration — Select from ElevenLabs' voice library with audio preview, adjust stability (controls consistency), and similarity boost (controls clarity)
- Behavior Settings — Define the system prompt, first message (greeting), temperature (response creativity), and language
- First Message — Set the initial greeting when conversations start
Step 5: Test Your Agent
Once configured, test your agent by asking it questions. For example, if you built a film noir detective agent, ask:
"Tell me about Casablanca."
Your agent should respond in character, with the voice and personality you configured.
Step 6: Edit or Duplicate Existing Agents
To edit an existing agent, provide the agent ID:
To duplicate an agent for a new market, update the voice and first message.
What You Can Build with ElevenLabs MCP
Voice Agents for Outbound Calls
Create voice agents to perform outbound calls — order a pizza, book an appointment, or follow up with a lead. The agent can speak in any voice you configure, with any personality you define through the system prompt.
Transcription and Audio Processing
Upload audio files and ask Claude to transcribe them, identify different speakers, or isolate background noise. The MCP server handles all audio processing tasks.
Voice Cloning and Custom Voices
Convert recordings to sound like different characters. Create custom voices by uploading samples, then use them in any agent or voiceover.
Sound Effects and Music
Generate soundscapes from text descriptions. For example: "Create a soundscape of a thunderstorm in a dense jungle with animals reacting to the weather."
Example Prompts for Voice Agent Building
"Create an AI agent that speaks like a film noir detective and can answer questions about classic movies."
"Generate three voice variations for a wise, ancient dragon character, then I will choose my favorite voice to add to my voice library."
"Convert this recording of my voice to sound like a medieval knight."
"Create a soundscape of a thunderstorm in a dense jungle with animals reacting to the weather."
"Turn this speech into text, identify different speakers, then convert it back using unique voices for each person."
"Create a customer support agent for my SaaS product with a friendly, helpful personality."
Advanced Configuration
File Output Modes
You can configure how the MCP server handles file outputs using the ELEVENLABS_MCP_OUTPUT_MODE environment variable:
- files — Save files to disk and return file paths (default)
- resources — Return files as MCP resources; binary data is base64-encoded
- both — Save files to disk AND return as MCP resources
Resource mode is useful for cloud environments or MCP clients without file system access, as files are returned directly in the MCP response.
Base Path Configuration
Set the base path for file operations with ELEVENLABS_MCP_BASE_PATH (default: ~/Desktop). This directory serves as the security boundary for input files.
Troubleshooting
- Logs — On macOS, logs are at ~/Library/Logs/Claude/mcp-server-elevenlabs.log. On Windows, they're at %APPDATA%Claudelogsmcp-server-elevenlabs.log.
- Timeout errors — Some ElevenLabs API operations like voice design and audio isolation can take a long time. This shouldn't occur when using a production client like Claude.
- API key issues — If you see authentication errors, confirm your API key is valid and has sufficient credits.
Frequently Asked Questions
Can I build voice agents without coding?
Yes. The ElevenLabs MCP server lets you create voice agents entirely through natural language prompts in Claude. No coding required. The interactive UI handles all the configuration.
What voices are available?
ElevenLabs offers thousands of voices across dozens of languages. You can browse the voice library, preview voices, and select the ones that fit your agent's personality.
Can I use my own voice?
Yes. You can upload voice samples to create custom voices, then use them in any agent or voiceover. The MCP server supports voice cloning directly from your chat.
Is ElevenLabs MCP free?
Yes. The free tier gives you 10,000 credits per month to test the MCP server and build voice agents.
Ready to Build Your First Voice Agent?
ElevenLabs MCP makes building AI voice agents accessible to everyone — no coding required. You can create conversational agents, generate speech, clone voices, and make outbound calls, all from your Claude chat.
Start with the free tier, set up the MCP server in minutes, and build your first voice agent today.
Try ElevenLabs MCP free today and start building AI voice agents from your chat.