# Audio MCP server

> Lets an AI client edit, analyze, and convert audio files using non-destructive manipulations and high-level plugins.

- Repository: https://github.com/audiojs/audio
- Page: https://hostedmcp.org/servers/audiojs/audio
- Transport: stdio
- Stars: 334
- License: MIT
- Latest version: 2.10.0

## Install

### Claude Code

```sh
claude mcp add audio -- npx -y audio
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "audio": {
      "command": "npx",
      "args": [
        "-y",
        "audio"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "audio": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "audio"
      ]
    }
  }
}
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "audio": {
      "command": "npx",
      "args": [
        "-y",
        "audio"
      ]
    }
  }
}
```

### Codex (`~/.codex/config.toml`)

```toml
[mcp_servers.audio]
command = "npx"
args = ["-y", "audio"]
```

### Gemini CLI (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "audio": {
      "command": "npx",
      "args": [
        "-y",
        "audio"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `ANTHROPIC_BASE_URL` | no | no |  |

## FAQ

### Can I handle very large audio files?

Yes, the paged architecture allows you to open files larger than 10GB without hitting the 2GB memory limit.

### Is the editing process destructive?

No, the server uses virtual edits and non-destructive processing, allowing for infinite undo and instant cloning.

### What kind of audio analysis can be performed?

It can analyze loudness, spectrum, beats, pitch, chords, and key.