LMMS Agent Bridge

This gives your AI agent access to LMMS so it can compose, arrange, mix, and export music, either working alone or alongside you in the same project.

The bridge connects any MCP-capable agent to a running LMMS session. The agent works inside the song that is open on your screen: it creates tracks, writes melodies and drum patterns, loads presets and samples, adjusts the mix, automates parameters, and renders audio. You can watch, play back, and edit in LMMS at the same time.

An arrangement composed and built by an agent through the bridge.

What your agent can do

• Compose: create instrument, sample, automation, and Beat/Bassline tracks; write and edit notes in batches; and place patterns across the song.

• Arrange: move, resize, duplicate, and rename clips; set the tempo and time signature.

• Sound design: find and load factory presets, samples, and VST instruments, then read and set instrument parameters.

• Mix: set volume, pan, mute, and solo; route tracks to mixer channels; manage effect chains; and write automation.

• Listen and export: play, pause, and seek; render WAV, FLAC, OGG, or MP3 audio (solo tracks first to render stems); and export MIDI. Each render reports its peak and RMS levels.

• Collaborate: open your existing songs, inspect what is already there, and build on it. Saving is under your control.

Related MCP server: openDAW MCP

How it works

AI agent  ──MCP (stdio)──▶  scripts/lmms_mcp.py  ──named pipe──▶  agentbridge.dll (inside LMMS)

Requirements

• Windows 10/11 x64

• LMMS 1.2.2 x64, installed to the default C:\Program Files\LMMS, or pass a different path to the launcher

• Python 3.10 or newer

• Git for Windows (provides Git Bash for the build)

• An MCP-capable agent or client

The setup script downloads the pinned MinGW compiler and the LMMS, Qt, and libsamplerate sources and verifies their SHA-256 hashes. Nothing is installed into the LMMS program directory.

Installation

Prebuilt binaries are not published yet, so build the plugin from source. Run these commands from the repository root.

1. Clone and create a Python environment (PowerShell):

git clone https://github.com/Jx9x/lmms-mcp-bridge.git lmms-agent-bridge
cd lmms-agent-bridge
py -3 -m venv .venv
& .\.venv\Scripts\python.exe -m pip install -r requirements.txt
& .\.venv\Scripts\python.exe scripts\setup_dependencies.py

2. Build the plugin (Git Bash):

./scripts/build-plugin.sh

This produces build/lmms-plugins/agentbridge.dll.

3. Launch LMMS with the bridge. Close any running LMMS first, then double-click start-lmms-with-bridge.bat, or run:

powershell -NoProfile -ExecutionPolicy Bypass -File scripts\start-lmms-with-bridge.ps1 -Project "C:\Music\my-song.mmp"

-Project is optional, and -LmmsExe selects a non-default LMMS installation. Launch LMMS this way whenever you want the agent to connect: starting LMMS normally does not load the plugin.

Connect your agent

1. Register the MCP server. Use absolute paths to the virtual environment's Python and to scripts/lmms_mcp.py. Most clients accept this JSON format; check your client's documentation for where to put it.

{
  "mcpServers": {
    "lmms": {
      "command": "C:/path/to/lmms-agent-bridge/.venv/Scripts/python.exe",
      "args": ["C:/path/to/lmms-agent-bridge/scripts/lmms_mcp.py"]
    }
  }
}

Optional arguments: --timeout <seconds>, --pipe <name> (it must match LMMS_AGENT_BRIDGE_PIPE in the LMMS environment), and --include-tool <name> for optional diagnostic tools.

2. Install the skill. Copy the .agents/skills/lmms-operation-instructions/ folder, including its references/ subfolder, into your agent's skills directory. If your agent does not support skills, give it SKILL.md as instructions. The skill does not need to be installed in the bridge's folder.

3. Try it. With LMMS running through the launcher, ask your agent:

Check the LMMS project info, then write an 8-bar lo-fi hip hop loop at 85 BPM with drums, bass, and an electric piano, and play it for me.

To test the connection without an agent:

& .\.venv\Scripts\python.exe scripts\bridge_client.py ping

Example requests

• "Load C:\Music\demo.mmp and add a countermelody to the second chorus."

• "The kick is buried. Balance the drums against the bass and render stems so I can compare them."

• "Write a four-chord progression in D minor on a pad, then arrange an intro, verse, chorus, and outro."

• "Export the song as a 24-bit WAV to C:\Music\exports\final.wav."

Troubleshooting

If a command times out or the connection drops, the outcome of that edit is unknown. Ask the agent to inspect the project before retrying; edits are never retried automatically.

Known limitations

• Only Windows x64 and LMMS 1.2.2 are supported. Newer LMMS releases (1.3/nightly) have a different plugin interface.

• VST loading has been verified only with SyndtSphere (VST2 x64). Other VSTs may load, but appearing in a VST scan does not confirm compatibility.

• Full ZynAddSubFX preset loading is not verified.

• Stock LMMS 1.2.2 does not save sample-clip names, so those names are lost when the project is reloaded.

• Project, preset, sample, and export paths must be absolute local drive paths. Network (UNC) and relative paths are rejected.

• Parameter IDs and track/clip handles are valid only for the current session.

Documentation

License

The native plugin, MCP server, and agent skill are licensed under the GNU General Public License v2.0 or later. The plugin compiles LMMS source code, which is itself licensed under GPL-2.0-or-later. See THIRD_PARTY_NOTICES.md for upstream components. The test fixture tests/fixtures/crunk-demo.mmp is from LMMS and is licensed under the Artistic License 2.0 (see tests/fixtures/LICENSES.TXT).

LMMS is a separate project and is not affiliated with this one.