Current packaged release: 1.2.5.12
Windows score editor and AI-assisted composition workstation.
Write, edit, arrange, and refine music in a full score editor, then use built-in AI tools or MCP-connected clients when you want generation help.
Park Byoung Gu (λ°λ³κ΅¬)
Director & SW Group Manager at AXT (Ajinextek) β a leading company in industrial motion controller markets β where he leads software R&D.
ArkComposer is a personal hobby project, built on weekends and spare moments, driven by a deep love of music.
- Full score editing and composition tools live in one workflow instead of separate notation and generator apps
Edit Measure (MIDI)gives you a modeless piano-roll editor for precise measure-level repair, timing edits, velocity shaping, and AI-assisted rewriting- Built-in composition helpers cover rhythm, melody variation, chord suggestion, voice leading, accompaniment, and AI generation
- Melody recording can capture microphone input or system sound playback and convert the detected melody into editable score notation
- An MCP server is built into the app, so external AI clients can control the same score editor when needed
- Visual analysis overlays and score-aware editing tools stay tied to the musical structure instead of raw MIDI only
- Undo / Redo stays at the far left for quick recovery
Edit MeasureopensEdit Measure (MIDI)for the current measure selectionMeasure Select,Note Select, andScore Editsplit selection from direct score editing- Pinned note and eraser tools can switch the app into score editing immediately
- Rhythm, Melody, Helper, and AI Helper menus are grouped on the right side of the top bar
- A status bar below the top menu shows the current mode, tool, track, selected measures, rhythm, melody, and AI state
- Multi-track score editor with treble and bass clef support
- Note input: whole, half, quarter, 8th, 16th, 32nd (+ dotted variants)
- Accidentals, dynamics (p, mf, f, cresc, dim), tempo marks (accel, rit)
- Key signature: major/minor, all 15 keys, 3 apply modes (pitch-fixed / position-fixed / transpose)
- Time signatures: 4/4, 3/4, 2/4, 6/8, 12/8
- Chord symbol display above measures
- Lyrics editor with per-measure editing
- Multi-document tabs with independent undo/redo per tab
- Piano-roll style editor for note timing, duration, pitch, and velocity inside selected measures
- Edit / Insert / Delete / Tie modes for fast repair without leaving the score workflow
- Track-wide measure navigation slider for moving across the active track while staying in the same editor
- Built-in playback controls for checking only the current track or all tracks
- Velocity popup editor for shaping dynamics inside the current measure range
- Designed for repeated measure cleanup and detailed note-level correction after generation or manual entry
- Add, delete, merge, split (by pitch range), reorder tracks
- Hide/show individual tracks
- Per-track instrument settings: SoundFont, Program (0β127), MIDI Channel, Bank MSB/LSB, Volume, Pan, Percussion
Rhythmreassigns rhythm patterns to selected measuresMelodyapplies transforms such as inversion and other note-sequence variations- Melody recording supports mic-in or sound playback capture, then converts the detected melody into editable score notes
Helpergroups chord suggestion, voice leading, accompaniment, and composition guidanceAI Helpergroups AI-backed generation tools for voicing and accompaniment- Voice Leading provides tunable parameters for harmonic line shaping and chord motion
- Accompaniment supports Arpeggio, Chord Beat, Drum Groove, and AI-driven styles
AI Voicingis available from the AI Helper menu and directly insideEdit Measure (MIDI)- Use
AI Voicingwhen the current track already has harmonic intent but needs cleaner voicing, redistribution, or regeneration AI AccompinsideEdit Measure (MIDI)generates accompaniment that follows the current melody and harmony while staying in the measure-edit workflow- AI accompaniment styles include
AI Chord Beat,AI Arpeggio,AI Free,AI Free (Melody Variation), andAI Drum Beat - Built-in ONNX and TinyGRU models can be selected from the app for local generation
- Full MCP server built into the app (stdio and TCP modes)
- Compatible with Claude Code and other MCP-capable clients
- Compose through natural language, then continue refining the result in the score editor or
Edit Measure (MIDI) - Scope control includes targeted edits such as selected range and whole-song operations
- MCP exposes self-documentation so connected AI clients can inspect available tools
- Playback with SoundFont synthesis (.sf2)
- Record melody from microphone input or system sound playback and convert it to editable score notation
- Export audio: WAV (lossless) / MP3 (compressed, requires
lame.exeinTools/) - Export score: PDF / PNG / JPG
- Export video: MP4/H.264 with score frames, playback cursor, and optional AAC audio via bundled FFmpeg
- Import / Export MIDI
- Native save format:
.arkusing ArkScore v3 compact JSON - Legacy score formats remain readable:
.json,.arkscore,.arkscore.json
- Built-in audio effect filters (Distortion, Echo, and more)
- Extensible plugin architecture for custom effects
Launch ArkComposer in MCP stdio mode:
ArkComposer.exe --mcp-modeAdd to your claude_desktop_config.json:
{
"mcpServers": {
"arkcomposer": {
"command": "C:/path/to/ArkComposer.exe",
"args": ["--mcp-mode"]
}
}
}Then talk to Claude Code:
"Create an 8-bar melody in C major, 4/4 time"
"Add an arpeggio accompaniment track"
"Apply jazz voice leading to measures 1β4"
"Continue the melody for 4 more measures"
- Launch ArkComposer normally
- In the left palette, click MCP Server
- Enter a TCP port (e.g.
7890) β Click Start - Connect your MCP-compatible client to
localhost:7890
Note on ChatGPT: ChatGPT does not natively support MCP.
To use ArkComposer with ChatGPT, you need to implement a separate OpenAPI server that bridges ChatGPT function calls to ArkComposer's MCP tools.
ArkComposer exposes a full score editing API over MCP:
| Category | Tools |
|---|---|
| Score reading | get_score, get_score_range, get_score_compact, get_score_range_compact |
| Note editing | add_note, add_notes_batch, add_notes_compact, change_note, delete_note |
| Track management | add_track, delete_track, set_track_props |
| Measure management | add_measure, delete_measure, delete_measures_range, clear_measure, set_measure_props |
| Document | new_song, set_title, undo |
Full AI usage guide:
docs/HowToUse_ArkCompoerForLLM.mdAlso available at runtime via MCP resource:arkcomposer://docs/how-to-use-llm
- Windows 10 / 11 (64-bit)
- MP3 export uses
Tools/lame.exe. - MP4 video export uses
Tools/ffmpeg.exe.
Place .sf2 files in the SoundFonts/ folder.
ArkComposer ships with:
GeneralUser GS v1.471.sf2β GeneralUser GS License v2.0 (S. Christian Collins)GeneralUser_GS_SoftSynth_v1.44.sf2β GeneralUser GS License v2.0 (S. Christian Collins)
π Latest Release
- macOS / Linux support (C++ / JUCE core is cross-platform ready)
- Cloud sync
- Plugin marketplace
- More AI model integrations
| Component | Author | License |
|---|---|---|
| JUCE Framework | Raw Material Software | JUCE Starter |
| GeneralUser GS v1.471.sf2 | S. Christian Collins | GeneralUser GS License v2.0 |
| GeneralUser_GS_SoftSynth_v1.44.sf2 | S. Christian Collins | GeneralUser GS License v2.0 |
| FluidSynth | Peter Hanappe and contributors | LGPL 2.1 |
| LAME MP3 Encoder | Mark Taylor and contributors | LGPL 2 |
| FFmpeg | FFmpeg developers / Gyan.dev Windows build | GPLv3 for the bundled static essentials build |
| Demucs / HTDemucs-6S ONNX model | Meta / Facebook Research; ONNX conversion artifact referenced by Tools/demucs/manifest.json |
Demucs upstream MIT; preserve model/source attribution and applicable artifact terms |
| ONNX Runtime | Microsoft and contributors | MIT |
| SDL3 | Sam Lantinga and contributors | zlib |
| libsndfile | Erik de Castro Lopo and contributors | LGPL 2.1 |
See Credits.txt for full details.
- FFmpeg is included as a separate
Tools/ffmpeg.exeexecutable for audio conversion/analysis workflows. The bundled binary reports a Gyan.dev essentials build configured with GPL/version3 options, so FFmpeg copyright, GPL terms, build/source availability, and upstream notices must be preserved when distributing ArkComposer. - Demucs / HTDemucs-6S is included as a local AI model asset for audio source-separation-assisted workflows. Keep the Demucs/Meta attribution, model artifact source information from
Tools/demucs/manifest.json, and any applicable model artifact license notice with redistribution. - ONNX Runtime is included as
onnxruntime.dllandonnxruntime_providers_shared.dllfor local ONNX inference. Preserve the Microsoft/contributor attribution and MIT license notice. - GeneralUser GS SoundFont files are included as bundled sound assets. Preserve the S. Christian Collins attribution and the GeneralUser GS License v2.0 notice.
- These third-party components are not owned by ArkComposer. Their licenses apply to their own binaries, libraries, models, and assets separately from the ArkComposer application license.
ArkComposer is free to download and use.
Music created with ArkComposer may be used for personal or commercial purposes without restriction.
Redistribution of ArkComposer itself is not permitted.
Source code is proprietary and not included in this distribution.
Built with C++ and JUCE
Website: arkcomposer.com
