The Audial MCP server puts Audial's audio tools inside your AI assistant. Install it once and Claude Desktop, Claude Code, Cursor or any other MCP client can split stems, analyze tempo and key, segment an arrangement, master a mix, build sample packs, convert audio to MIDI, generate music, resynthesize a one-shot into a synth preset, and sing lyrics in a reference voice.
It is a Python package on PyPI called audial-mcp and it runs locally with uvx audial-mcp, so there is nothing to clone or build. Configuration is two environment variables, AUDIAL_API_KEY and AUDIAL_USER_ID, taken from your Audial dashboard. Credentials in your client config always win: the server snapshots its environment before the Audial SDK can read any config file on disk.
Tool calls write their output to disk instead of returning audio inline, so your assistant passes file paths around rather than megabytes of encoded audio. Results land in a per-tool, timestamped folder under a central ~/Audial folder, which you can move with AUDIAL_RESULTS_DIR, and list_results browses them without leaving the chat. The server is listed in the official MCP registry as io.github.AudialAI/audial-mcp.
What you get
- Ten tools: stem_split, analyze, segment, master, generate_samples, generate_midi, generate_music, sound2vital, text2vox and list_results
- Installs with uvx audial-mcp, no clone and no build step
- Works with Claude Desktop, Claude Code, Cursor and any client that supports stdio MCP servers
- Results saved to a central ~/Audial folder and browsable with list_results
- Listed in the official MCP registry as io.github.AudialAI/audial-mcp
- Configured with AUDIAL_API_KEY and AUDIAL_USER_ID from your dashboard
Use it from code
# Claude Code
claude mcp add audial \
-e AUDIAL_USER_ID=your-user-id -e AUDIAL_API_KEY=your-api-key -e AUDIAL_RESULTS_DIR=~/Audial \
-- uvx audial-mcp
# Then ask: "Split ~/Music/song.mp3 into stems and master the result."// claude_desktop_config.json (Claude Desktop), ~/.cursor/mcp.json (Cursor),
// or the equivalent config for any other stdio MCP client
{
"mcpServers": {
"audial": {
"command": "uvx",
"args": ["audial-mcp"],
"env": {
"AUDIAL_USER_ID": "your-user-id",
"AUDIAL_API_KEY": "your-api-key",
"AUDIAL_RESULTS_DIR": "~/Audial"
}
}
}
}# The MCP tools wrap the same Python SDK you can call directly
import audial
audial.config.set_api_key("your_api_key")
audial.config.set_user_id("your_user_id")
result = audial.stem_split("song.mp3")
print(result["files"]["folder"])Frequently asked questions
Install uv, then add the server to your client. In Claude Code it is one claude mcp add command; in Claude Desktop and Cursor it is a JSON block that runs uvx audial-mcp. The Audial plugin for Claude does the same setup in one step.
Ten: stem_split, analyze, segment, master, generate_samples, generate_midi, generate_music, sound2vital, text2vox and list_results. Every tool except list_results calls the Audial API.
Into a per-tool, timestamped folder under your results directory, which defaults to ~/Audial. Set AUDIAL_RESULTS_DIR to move it, and use list_results to browse past jobs.
Claude Desktop, Claude Code and Cursor have documented setups, and any client that supports stdio MCP servers can use the same JSON configuration.
The audio and text you pass to a tool are uploaded to https://api.audialmusic.ai over HTTPS and processed on Audial's servers. Results are downloaded back into your results folder. Nothing else is sent.
Yes for every tool except list_results. Those calls hit the Audial API, which answers HTTP 402 SUBSCRIPTION_REQUIRED without an active subscription on your account.