Text to Speech MCP Server: Let Your AI Make Voice-Overs in Your Clonylab Tab

Start with the plain version. MCP stands for Model Context Protocol. It is an open standard that gives AI apps one common way to use outside tools. Think of a standard plug. Once an app has the plug, any tool with the matching socket works, and nobody has to build a custom connection for every pair of apps.

A text to speech MCP server is the tool on the other end of that plug. It offers actions such as “list the voices”, “turn this text into speech” or “show me my last voice-over”. Your AI reads that menu, decides which action fits your request, and calls it for you. People also search for the same thing as a TTS MCP server, an MCP text to speech server, MCP voice generation or an AI voice MCP.

Here is a small scene. You ask your AI to write a 60-second script for a product video. Normally you would copy the script, open a voice tool, paste it, choose a voice, press a button and download the file. With a text to speech MCP server you add one more sentence to the same request: “and make the voice-over in a calm US voice.” The AI does the rest.

Clonylab runs one of these servers. It lives at a single address, https://clonylab.com/mcp, and the address is the same for every AI app. If the idea is new to you, the official MCP site has a short introduction, linked at the end of this guide.

A voice-over is easy to make once. It gets tiring when you make one every week. The slow part is rarely the audio itself. It is the hand-offs: moving text from one app to another, picking settings, saving files and finding them again later.

MCP removes most of those hand-offs. A few jobs where that helps:

• Video creators. Draft the script and get the narration in the same chat.
• Course and training teams. Ask for one lesson script after another and collect the MP3 files as you go.
• Developers. Ask your editor to narrate a code walkthrough or a changelog.
• Product and support teams. Keep a set of how-to clips that your AI can make again when the text changes.

The chart counts what you do yourself in each way of working. It is our own count for one voice-over, not a timing test, and your own numbers will differ.

Most text to speech services run on a server and send you the audio. Clonylab works differently, and that shapes how its MCP server behaves.

Your AI app sends a request to the MCP address. Our server checks who you are, then passes the command to your open Clonylab tab. The tab makes the voice-over with the same code the Studio uses and saves it to your History, which lives in your browser. When your AI asks for the finished file, the tab sends the MP3 back and your AI gets a download link.

That is why one rule matters: keep Clonylab open in a browser tab while your AI works. If the tab is closed, the tools that make audio or read your History cannot run. A tool called connection_status lets your AI check this before it starts.

Every speech job has a status: queued, downloading, generating, done, error or cancelled. A long script does not make your AI wait. It gets a job ID straight away and can ask for the status later.

The server is not a black box. These are the real tools your AI sees, grouped by what they do. If you have used the Studio, every one of them will feel familiar.

generate_speech takes up to 10,000 characters of text per call. It understands the same script tags as the Studio: [pause 1s], [slow], [fast], [soft], [loud] and [VoiceName] to switch voice partway through. Cloned voices use the ids from list_cloned_voices and work in English only.

get_audio returns a link you can download with a command such as curl. Clips under 1 MB are also attached as audio, so your AI app may be able to play them right in the chat.

The Connector screen in your account has every value you need, already filled in. The steps below follow that screen.

Get a plan that includes the AI connection

The AI connection is part of Pro and Team. The free trial lets you try the Studio and cloning first, so you can hear the voices before you upgrade.

Copy your MCP address

Open the Connector screen from the side menu. The first card shows the address. It is the same for every AI tool.

clonylab.com/connect

Connect your AI app

There are two ways, and the screen shows both.

Option A: sign in. For apps with an “add custom connector” screen, such as Claude and ChatGPT. Paste the address, leave any client ID or secret empty, choose “Sign in”, and press Allow on the Clonylab page that opens. In Claude, choose “Register automatically”.

clonylab.com/connect

Option B: a token. For tools you set up in a terminal or a config file, such as Claude Code, Cursor, Codex and Gemini CLI. Give the token a name, such as the tool and the computer, and copy it. It is shown only once.

clonylab.com/connect

For Claude Code, one command does it:

Claude Code (run once in your terminal)Copy

claude mcp add --transport http clonylab https://clonylab.com/mcp --header "Authorization: Bearer YOUR_TOKEN"

Cursor reads a short JSON file. Put this in ~/.cursor/mcp.json:

Cursor (~/.cursor/mcp.json)Copy

{
  "mcpServers": {
    "clonylab": {
      "url": "https://clonylab.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

clonylab.com/connect

Claude Desktop needs Node.js, because it uses a small helper called mcp-remote. Any other tool that supports MCP only needs two values: choose “Streamable HTTP” and enter the address and the header Authorization: Bearer YOUR_TOKEN.

Let your AI use this browser

This switch is off by default. Turn it on and keep Clonylab open in a tab while your AI works.

clonylab.com/connect

You do not need special words. A good prompt names the text, the voice or the accent, and what you want back. Here are five that map to real tools.

clonylab.com/connect

Here is what the server needs from you, and what it cannot do. We would rather tell you now than surprise you later.

Giving an AI app access to a tool deserves a careful look. Clonylab keeps the controls in your hands.

• You approve the connection. With sign-in, a Clonylab page asks you to press Allow. With a token, you create it yourself.
• You can undo it. Remove an app from the list or delete a token, and that tool stops working.
• It is off by default. The browser switch stays off until you turn it on, and you can turn it off again whenever you like.
• Your own text is treated as data. The History tools tell your AI that saved scripts are data and not instructions, which helps against a script that tries to give orders.

Your AI tool sends commands through our server to your tab. The audio is made and stored in your browser, and it goes back to the AI only when the AI asks for a finished voice-over. Read the privacy policy for the full picture.

You do not have to pick one. The Studio is great for a single voice-over where you want to listen and adjust. The MCP server shines when your AI is already part of the job.

Not sure which voices to try first? The standard voices page lets you hear all 38. The AI voiceover generator page explains the script tags in detail.