ElevenLabs MCP in Claude means connecting ElevenLabs' voice API directly to Claude Desktop or Claude Code through the Model Context Protocol, so you can ask Claude in plain English to generate speech, clone a voice, transcribe audio, or design a sound effect without touching a line of API code. Setup takes about five minutes once you have an API key and the uv package manager installed, and after that Claude just gains a new set of tools it can call on its own.
Short answer: Install
uv, grab a free ElevenLabs API key, then add a short JSON block to Claude Desktop's config (or run oneclaude mcp addcommand in Claude Code) pointing at theelevenlabs-mcppackage. Restart Claude and it gains tools for text-to-speech, voice cloning, transcription, and sound design — no API code required, free plan included.
I set this up on a clean machine for this guide — fresh Claude Desktop install, fresh ElevenLabs account, no leftover config from a previous project — specifically so I'd hit whatever a first-time user hits. It took under ten minutes end to end, including the one step that trips people up (more on that below). Here's the setup path, the prompts that actually get useful output, and where I got stuck the first time through.
What you'll need
An ElevenLabs account and an API key from elevenlabs.io/app/settings/api-keys — the free plan works fine to start, and per ElevenLabs’ own pricing page it covers 10,000 credits a month, roughly ten minutes of generated speech. You'll also need uv, the Python package manager ElevenLabs' official server runs on; install it with one terminal command on Mac or Linux, or the equivalent installer on Windows. Claude Desktop or Claude Code both work — Desktop uses a config file, Code uses a CLI command, and I tested both. No coding is required for either path, but you do need to be comfortable editing a JSON file or running one terminal command. Budget five to ten minutes total.
Step-by-step: ElevenLabs MCP in Claude
1. Get your ElevenLabs API key
Log into ElevenLabs, go to Settings → API Keys, and generate a new key. Copy it somewhere safe — you'll paste it into a config file in a minute, and ElevenLabs won't show you the full key again after you navigate away.
2. Install uv
Open a terminal and run curl -LsSf https://astral.sh/uv/install.sh | sh on Mac or Linux. On Windows, use the PowerShell installer linked from the uv project's GitHub page. This is the piece people skip, assuming Claude will handle it — it won't, and per ElevenLabs’ official MCP server README, uv is a hard prerequisite, so the server fails silently without it.
3. Edit Claude Desktop's config file
In Claude Desktop, go to Claude → Settings → Developer → Edit Config, which opens claude_desktop_config.json in your default text editor. Add this block, swapping in your real key where it says your-api-key-here:
“json { "mcpServers": { "ElevenLabs": { "command": "uvx", "args": ["elevenlabs-mcp"], "env": { "ELEVENLABS_API_KEY": "your-api-key-here" } } } } “
If you're on Windows, also enable Developer Mode first — hamburger menu → Help → Enable Developer Mode — or the config change won't take effect.
4. Restart Claude Desktop and confirm the connection
Quit Claude completely and reopen it, not just close the window. Start a new chat and check the tools icon near the message box — ElevenLabs should now appear as a connected server with its tools listed underneath.
5. Or connect it to Claude Code instead
If you live in the terminal, skip the JSON editing entirely and run:
“bash claude mcp add --env ELEVENLABS_API_KEY=your-api-key-here --transport stdio elevenlabs -- uvx elevenlabs-mcp “
That's the whole setup for Claude Code — one command, no file to hunt down. Per Claude’s own MCP documentation, add --scope user on the end if you want ElevenLabs available in every project instead of just the current one.
6. Set an output folder (optional but worth doing)
By default, generated audio lands on your Desktop, which gets messy fast. Add "ELEVENLABS_MCP_BASE_PATH": "/path/to/folder" inside the same env block to redirect output somewhere sane.
Example prompts you can copy
These are close to what I actually typed while testing the setup above:
- "Turn this paragraph into speech using a calm, professional voice: [paste text]."
- "Clone my voice from this 30-second sample and read this script back in it: [attach audio file]."
- "Transcribe this interview recording and flag any spot where the speaker sounds unsure or trails off: [attach audio file]."
- "Generate a 10-second sound effect for a door creaking open in an empty house."
- "Take this podcast intro script and read it in three different voice styles so I can pick one: [paste script]."
Naming the format and the specific outcome — "calm, professional voice," "flag where the speaker sounds unsure" — gets a usable result on the first try. Just saying "make this into audio" gets you ElevenLabs' generic default voice, which is rarely what you actually want.
Common mistakes to avoid
The mistake that cost me the most time was skipping the uv install and assuming uvx was already on my system — it wasn't, and Claude Desktop just showed the ElevenLabs server as disconnected with no obvious error message pointing at the real cause. Second, forgetting to fully quit and relaunch Claude after editing the config; a simple window close doesn't reload it. Third, pasting the API key with a stray space or quote mark left over from copying it out of the ElevenLabs dashboard — the JSON parses fine but the key silently fails auth. Fourth, not checking your ElevenLabs plan's credit balance before a big batch job; the free tier's 10,000 credits disappear fast if you're cloning voices or generating long scripts, and the MCP call just returns an error once you're out. Fifth, treating every generation as final — in my testing, the second pass with a specific note ("less robotic," "slower pacing") consistently beat regenerating blind.
ElevenLabs plans compared
| Plan | Monthly price | Credits/month | Voice cloning |
|---|---|---|---|
| Free | $0 | 10,000 (~10 min audio) | Not included |
| Starter | $6 ($5/mo annual) | 30,000 | Instant voice cloning |
| Creator | $22 ($18.33/mo annual) | 121,000 | Professional voice cloning |
| Pro | $99 ($82.50/mo annual) | 600,000 | Professional, higher-quality audio output |
| Scale | $299 ($249.17/mo annual) | 1,800,000 | 3 professional voice clones, team seats |
Prices and credit allotments are current as of ElevenLabs' own pricing page, checked August 18, 2026. Any of these plans works with the MCP server — the plan only controls how many credits you burn through and whether voice cloning is unlocked, not whether the Claude connection itself works.
Tools that make this easier
If you're new to Claude generally and want the full picture before layering MCP servers on top, how to use Claude AI covers accounts, plans, and Projects from scratch. ElevenLabs isn't the only voice option worth knowing — if you mainly want spoken conversations rather than generated audio files, how to use ChatGPT voice is the comparable feature on the other side. For music instead of speech, my how to use Suno AI guide covers the tool built specifically for that. If MCP servers in general are new to you, my AI coding assistant guide explains the broader category of tools Claude can now plug into, and ChatGPT alternatives for coding is useful if you're deciding which assistant should anchor your MCP setup in the first place. And if Claude versus ChatGPT is still an open question for you, Claude vs ChatGPT compares both directly.
My take
Setting this up once versus building a custom ElevenLabs integration by hand is not a close call — the MCP server does the plumbing, and after the first restart it just works as a normal part of chatting with Claude. The one place I'd actually pay attention is credit usage: the free plan's 10,000 credits are fine for testing prompts and small clips, but if you're planning to generate audio regularly, budget for at least Starter or Creator before you get surprised by a mid-project error. Voice cloning specifically only unlocks on Starter and up, so check that before you promise a client instant cloning on a free account.
Frequently Asked Questions
Is ElevenLabs MCP in Claude free to use?
Yes, to start. The ElevenLabs free plan includes 10,000 credits a month and works with the MCP server, no card required. You'll need a paid plan once you exceed that usage or want voice cloning, which starts on the $6/month Starter tier.
How long does it take to set up ElevenLabs MCP in Claude?
About five to ten minutes if you already have an ElevenLabs account. Installing uv, editing the config file (or running the one-line Claude Code command), and restarting Claude is the entire process — no code to write.
What is the easiest way to connect ElevenLabs to Claude?
Claude Code's CLI is the fastest path: one claude mcp add command with your API key, no file editing. Claude Desktop takes one extra step — opening and editing claude_desktop_config.json by hand.
Does this work with Claude Code, or only Claude Desktop?
Both. Desktop uses a JSON config block; Code uses the claude mcp add CLI command shown above. The underlying elevenlabs-mcp package and its tools are identical either way.
What can I actually do with ElevenLabs MCP once it's connected?
Ask Claude to generate speech from text, clone a voice from a sample, transcribe audio files, run speech-to-speech voice conversion, isolate vocals from background noise, or generate short sound effects — all through normal chat, without opening ElevenLabs' own app.