MCP Server Guide

You don't actually need to install anything — the easiest path is to let your MCP client run it via npx. If you prefer a global install:

npm install -g stemsplit-mcp

1. Log in to StemSplit Settings
2. Open the API Keys section
3. Click Create New Key, name it "MCP Client", and copy the sk_live_… value

Pick the client you use. The config is the same shape everywhere — point at npx -y stemsplit-mcp and set STEMSPLIT_API_KEY in the env.

{
  "mcpServers": {
    "stemsplit": {
      "command": "npx",
      "args": ["-y", "stemsplit-mcp"],
      "env": {
        "STEMSPLIT_API_KEY": "sk_live_..."
      }
    }
  }
}

{
  "mcpServers": {
    "stemsplit": {
      "command": "npx",
      "args": ["-y", "stemsplit-mcp"],
      "env": {
        "STEMSPLIT_API_KEY": "sk_live_..."
      }
    }
  }
}

{
  "assistant": {
    "mcp_servers": {
      "stemsplit": {
        "command": "npx",
        "args": ["-y", "stemsplit-mcp"],
        "env": { "STEMSPLIT_API_KEY": "sk_live_..." }
      }
    }
  }
}

Open a new chat in your MCP client and try one of these prompts:

Available tools

The server also ships 5 resources (live balance, recent jobs, job detail, YouTube job detail, SoundCloud job detail) and 6 prompts (karaoke, isolate_dialogue, sampler_pack, youtube_instrumental, soundcloud_instrumental, clean_voice). See the README for the full reference.

separate_youtube takes the same optional fields but uses youtubeUrl instead of source, and output is fixed to vocals + instrumental at BEST quality. separate_soundcloud takes soundcloudUrl instead, with the same output constraints (max 15 min).

Use separate_soundcloud (or the /soundcloud_instrumental slash command) to extract vocals and an instrumental from any public SoundCloud track — no download needed.

{
  "soundcloudUrl": "https://soundcloud.com/artist/track-name"
}

Accepted URL formats

Use clean_voice to remove background noise, hum, hiss, echo, and ambient sound from any audio file. Powered by DeepFilterNet — works great on podcasts, voice-overs, dialogue recordings, and isolated vocals.

{
  "source": "/Users/me/Recordings/podcast-ep12.mp3",
  "outputFormat": "WAV"
}

clean_voice parameters

When wait=false, the tool returns a job_id immediately — use get_denoise_job to poll status and retrieve the download URL.

All configuration is done via environment variables in your MCP client config.