Audio Separation Server

Streamable MCP server for ComfyUI MelBandRoFormer audio source separation.

Tools

Tool	Description
audiosep_health	Check ComfyUI backend status
audiosep_list_models	List available MelBandRoFormer models
audiosep_start_job	Submit audio file for separation
audiosep_get_job	Get job status/results
audiosep_cancel_job	Cancel a running job
audiosep_list_artifacts	List all stored artifacts
audiosep_get_artifact	Download artifact by name

Running

MCP Server (streamable-http)

cd /home/tyrel/projects/audio-sep-mcp
AUDIOSEP_MCP_PORT=8769 python mcp_server.py  # serves on http://localhost:8769/mcp

Note: default port 8765 collides with other local MCP servers — the systemd service below overrides it to 8769.

With Gradio UI

python app.py  # serves on :8767

systemd (user services)

Both services are managed as systemd user units and start automatically (linger enabled):

systemctl --user status audio-sep-mcp.service audio-sep-gradio.service
systemctl --user restart audio-sep-mcp.service audio-sep-gradio.service
journalctl --user -u audio-sep-mcp.service -f

Unit files: ~/.config/systemd/user/audio-sep-mcp.service, ~/.config/systemd/user/audio-sep-gradio.service.

Environment Variables

Variable	Default	Description
COMFYUI_URL	http://localhost:8188	ComfyUI backend URL
AUDIOSEP_MCP_PORT	8765	MCP server port (systemd service overrides to 8769)
AUDIOSEP_WORKSPACE	~/Documents/audio-sep-output	Working directory
AUDIOSEP_WORKFLOW_PATH	~/ComfyUI/user/default/workflows/hermes_known_good/melband_roformer_vocals_instruments_preview.json	Workflow file
AUDIOSEP_OUTPUT_RETENTION_DAYS	30	Artifact cleanup age

Supported Input Formats

wav, mp3, flac, ogg, aac, m4a, mp4, webm

Output

Splits audio into vocals.wav and instruments.wav in ~/Documents/audio-sep-output/artifacts/<job_id>/.