Have Claude narrate your blog in your own voice (local MCP)

Here is a small thing that feels like the future: you finish writing a post, you ask Claude to read it back in your own voice, and a minute later there is an audio file on your Mac. No upload, no web dashboard, no copy and paste into a separate app. Claude does it by calling Vocast through a local MCP server. This post shows how that works and how to set it up.

#What the MCP server actually is

The Model Context Protocol (MCP) is a standard way for an agent like Claude to call tools that live on your machine. Vocast ships a local MCP server that speaks MCP over stdio, so it runs entirely on your Mac with no network in the loop. Once it is registered, Claude can see a handful of Vocast tools and call them the same way it calls any other tool.

The tools it exposes are small and specific:

• list_voice_profiles: the voices you have already built, so Claude can pick one.
• clone_voice: narrate a script in one of your voices, and return an audio file.
• denoise: clean a noisy recording.
• health: check what the engine can do on this machine.
• list_history: recent jobs.

There is no separate narrate tool: narration is what clone_voice does. You hand it text and a voice, and it renders the audio.

#Registering Vocast with Claude

Vocast prints the exact registration snippet for your install, so you never have to guess paths. For Claude Desktop, add a vocast entry under mcpServers in the Claude config file. It looks like this, with the command and args pointing at the runtime and server that ship inside Vocast:

{
  "mcpServers": {
    "vocast": {
      "command": "/path/to/vocast/python",
      "args": ["/path/to/vocast/mcp_server.py"]
    }
  }
}

For Claude Code, the one-liner is claude mcp add vocast -- /path/to/vocast/python /path/to/vocast/mcp_server.py, again with the exact paths Vocast gives you. Restart the client and Claude will list the Vocast tools.

#Narrating a post

Now the useful part. Point Claude at the text and ask, in plain language:

You: Narrate my latest blog post in my own voice and save the audio.

Claude:
  1. calls list_voice_profiles  -> finds "MyVoice"
  2. calls clone_voice(text="...the post...", profile_id="MyVoice")
  3. returns { output_path, pns }

Claude reads your voice profiles, renders the script in the one you name, and hands back a path to a wav file plus a prosody score. The whole render happens on the Mac, so the script and the audio never leave the machine. Because it is a normal tool call, you can chain it: clean a recording with denoise, then narrate a correction with clone_voice, all in one conversation.

#A few honest notes

• Voice cloning needs Apple Silicon. The clone_voice tool renders on the neural engine, so it runs on an M-series Mac. denoise is lighter and less picky.
• You need a profile first. Build one in the Vocast app from a short sample, then Claude can call it by name.
• It is synchronous. A long script can take a little while, since the audio is really being rendered, not streamed from a server.

That is the whole idea behind Vocast being AI-native: your own voice, as a tool any MCP-capable agent can call, running locally on your Mac. The narration in your voice is one prompt away.