AudioLasso ships a stdio MCP server in the audiolasso npm package.

Use it when an agent runtime supports MCP and should call AudioLasso as tools instead of writing raw HTTP.

Requirements

Node.js 20 or newer.
An AudioLasso API key in AUDIOLASSO_API_KEY.
An MCP client that can launch stdio servers.

Configure an MCP client

Use npx so the client can launch the package directly:

{
  "mcpServers": {
    "audiolasso": {
      "command": "npx",
      "args": ["-y", "audiolasso", "mcp"],
      "env": {
        "AUDIOLASSO_API_KEY": "al_live_..."
      }
    }
  }
}

For local development from this repository, build first:

pnpm package:build
AUDIOLASSO_API_KEY="al_live_..." node packages/audiolasso/dist/mcp.js

Tools

The MCP server exposes:

audiolasso_upload_file
audiolasso_submit_audio_separation
audiolasso_get_status
audiolasso_wait_for_result
audiolasso_get_result
audiolasso_list_models

Submit a public URL

Ask the agent to call audiolasso_submit_audio_separation with:

{
  "audioUrl": "https://example.com/audio.wav",
  "prompt": "isolate the lead vocal"
}

Then call audiolasso_wait_for_result with the returned request_id.

Submit a local file

The simplest local-file path is one tool call:

{
  "localPath": "./song.wav",
  "prompt": "isolate the vocals"
}

The server uploads the file first, then submits the separation job.

You can also call audiolasso_upload_file yourself and submit the returned file_id.

Notes for agents

Do not write secrets into prompts, docs, or skill files.
Use AUDIOLASSO_BASE_URL only for development or staging.
Return both data.target.url and data.residual.url when a job completes.
Use llms.txt as the compact discovery index for docs and API references.

MCP server