able-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@able-mcp What's the tempo and track count of my latest project?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ABLE-MCP
Right-click MIDI and audio tools for Ableton Live 12, plus an MCP server for .als analysis and live bridge orchestration.
Two artifacts in one repo:
AbleMCP.ablx — a native Ableton Live 12 extension. Drag it in, right-click clips, get fast deterministic transformations and vocal-to-complement generation. No API key, no cloud dependency.
able-mcp(Python) — an MCP server for parsing and analyzing.alsproject files offline. Plug into Claude Desktop / VS Code / Cursor and ask questions about your sets.
The extension (shipping in v0.2)
Built on @ableton-extensions/sdk 1.0.0-beta.0. Runs inside Live's own process — not Max for Live. Tested on Live 12.4.5b3.
Install
Download
dist/AbleMCP.ablxOpen Live → Settings → Library → Extensions
Drag the
.ablxfile inRestart Live
What you get
Right-click any MIDI clip (session or arrangement) → ABLE-MCP menu:
Action | What it does |
ABLE-MCP: Variation | 30% chance per note: bump ±octave; 15% chance: drop the note. |
ABLE-MCP: Octave double | Add a quieter octave-up layer. |
ABLE-MCP: Humanize | Jitter timing ±0.05 beats and velocity ±10. |
ABLE-MCP: Reverse | Mirror notes around the clip's midpoint in time. |
ABLE-MCP: Invert (mirror) | Mirror pitch around the median note. |
ABLE-MCP: Transpose +1 / ABLE-MCP: Transpose -1 | Semitone up / down. |
ABLE-MCP: Legato | Extend each pitch's note to the next one's start. |
ABLE-MCP: Staccato | Cut every note to 25% of its length. |
ABLE-MCP: Thin (every other) | Drop every other note in time order. |
ABLE-MCP: Double-time | Compress timing 2× and play the pattern twice. |
ABLE-MCP: Half-time | Stretch to 2×; first half plays at half speed. |
ABLE-MCP: Strum chords | Stagger simultaneous notes by 1/64 each (instant strum). |
ABLE-MCP: Accent downbeats | Boost on-beat notes, duck off-beat. |
ABLE-MCP: Top voice only | Keep the highest note of each chord. |
ABLE-MCP: Bass voice (drop 8va) | Keep the lowest note of each chord, drop an octave. |
ABLE-MCP: Shift… | Modal: shift notes by N beats and/or N semitones, with optional wrap-around. Quick-bump preset buttons for common values. |
Every action also works on time-range selections in arrangement view — right-click a selection across one or more MIDI tracks and pick ABLE-MCP (range): …. Only notes inside the range are transformed.
Right-click any audio clip in arrangement view:
Action | What it does |
ABLE-MCP: Chop into beats (1/4 bar) / 1/2 / 1 / 2 / 4 / 8 bars | Splits a long audio clip into adjacent cropped chunks without rewriting source audio. |
ABLE-MCP: Snap first hit to bar | Detects first onset in WAV and repositions crop/start so the first hit lands on a downbeat. |
ABLE-MCP: Create complementary MIDI from vocal... | Renders the clip range, extracts guide melody from audio, generates a complementary line, and writes a new arrangement MIDI clip (modal controls: similarity, density, register, call/response, seed, target track). |
Every action is a single Cmd-Z undo.
Why it's fast
Core MIDI/audio actions are pure deterministic functions running in-process. Typical transform actions are near-instant and fully undoable. No LLM call is required for the shipping extension feature set.
Building from source
cd extension
npm install
npm run build # tsc + esbuild → dist/extension.cjs
npm run package # → ../dist/AbleMCP.ablxRelated MCP server: io.github.peterkolbe/ableton-for-ai
The MCP server
Python 3.11, stdio transport. Parses Live's gzipped .als files and exposes structure as MCP tools. Useful when you want an LLM to read a project without opening Live.
Install
Option A — local clone:
uv syncOption B — run directly from GitHub (no PyPI required):
uvx --from git+https://github.com/bassDaddyDevices/ABLE-MCP@v0.2.0 able-mcpRun
Local clone:
uv run able-mcpGitHub source (one-shot):
uvx --from git+https://github.com/bassDaddyDevices/ABLE-MCP@v0.2.0 able-mcpHook into a client
Claude Desktop (local clone) — ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"able-mcp": {
"command": "uv",
"args": ["--directory", "/absolute/path/to/ABLE-MCP", "run", "able-mcp"]
}
}
}Claude Desktop (GitHub source, no PyPI):
{
"mcpServers": {
"able-mcp": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/bassDaddyDevices/ABLE-MCP@v0.2.0",
"able-mcp"
]
}
}
}VS Code (Copilot Chat, local clone) — your user or workspace mcp.json:
{
"servers": {
"able-mcp": {
"command": "uv",
"args": ["--directory", "/absolute/path/to/ABLE-MCP", "run", "able-mcp"]
}
}
}VS Code (Copilot Chat, GitHub source, no PyPI):
{
"servers": {
"able-mcp": {
"command": "uvx",
"args": [
"--from",
"git+https://github.com/bassDaddyDevices/ABLE-MCP@v0.2.0",
"able-mcp"
]
}
}
}Tools
Tool | What it does |
| Tempo, time sig, track/clip/note counts, master-bus state. |
| Per-track kind, name, mute/solo, device & clip counts. |
| Clip metadata across one or all tracks. |
| Note-level MIDI dump. |
| Heuristic suggestions for things that look incomplete. |
| Extract monophonic guide melody from vocal WAV. |
| Generate complementary (not duplicated) MIDI line. |
| One-shot extraction + complement generation pipeline. |
| Replace notes in arrangement MIDI clip by index. |
| Render audio range in Live, generate complement, create/write arrangement MIDI clip in one call. |
Roadmap
v0.2.x — tighten vocal-to-complement quality (better vibrato handling, stronger anti-unison controls, warped-clip timing support).
v0.3 — richer bridge orchestration tools (multi-clip workflows and arrangement/session interoperability).
v0.4 — expand deterministic composition primitives (scale quantize, swing, density, stutter, chord-tone targeting).
Develop
uv run pytest # Python tests
uv run ruff check . # Python lintPython tests use synthetic .als fixtures (gzipped XML) generated in tests/fixtures/init.py. When a real Live 12 file becomes available, drop it under tests/fixtures/real/ and add parity tests — the parser is best-effort against an undocumented schema.
Supply-chain hygiene
uv.lockis committed; review every diff.All Python deps pinned with
==.The extension has exactly one runtime dep (
@ableton-extensions/sdk), pinned in extension/package.json.
License
MIT (extension and Python server).
Available Tools
40 toolsals_extract_midiB
Extract MIDI notes from one clip (by track index and clip index) as pitch/start/duration/velocity tuples.
Inputs: path (required), clip_index (required), track_index (required). Returns: object
als_find_unfinishedC
Heuristic suggestions about what looks unfinished in a set: missing instruments, empty MIDI clips, no master limiter, etc.
Inputs: path (required). Returns: result
als_list_clipsA
List clips on one track (by index) or every track (omit track_index).
Inputs: path (required), track_index. Returns: result
als_list_tracksB
List all tracks in an .als file with their kind, name, mute/solo state, and device count.
Inputs: path (required). Returns: result
als_summaryA
Summarize an Ableton .als file: tempo, time signature, track and clip counts.
Inputs: path (required). Returns: object
assist_finishingC
Aggregate finishing suggestions for a set: missing instruments, empty clips, no master limiter, single-note placeholders, very-short clips.
Inputs: path (required). Returns: result
assist_harmonic_overviewA
Run key estimation and chord guessing across every MIDI clip in the set. Useful for a 'where am I harmonically' overview.
Inputs: path (required). Returns: object
live_arrangement_create_midi_clipC
Create an empty MIDI clip in the arrangement timeline of a MIDI track.
Inputs: duration (required), start_time (required), track_index (required), name, +1 more. Returns: object
live_arrangement_set_clip_notesB
Replace notes in an arrangement MIDI clip. notes is a list of dicts with {pitch, start, duration, velocity, mute?}.
Inputs: notes (required), track_index (required), arrangement_index (required), confirm. Returns: object
live_check_bridgeA
Check whether the AbleMCP M4L bridge is running. Returns the bridge's ping payload on success, or guidance on how to install the device.
Inputs: no named parameters. Returns: object
live_continueC
Continue playback from the current position.
Inputs: confirm. Returns: object
live_create_clipB
Create an empty MIDI clip in the given session slot. length_beats defaults to 4.
Inputs: scene_index (required), track_index (required), name, confirm, +1 more. Returns: object
live_create_sceneC
Create a new scene. index defaults to -1 (append).
Inputs: index, confirm. Returns: object
live_create_trackA
Create a new track. kind ∈ {'midi','audio','return'}; index defaults to the end. Returns the new track info.
Inputs: kind, index, confirm. Returns: object
live_delete_clipC
Delete a clip from a session slot.
Inputs: scene_index (required), track_index (required), confirm. Returns: object
live_delete_trackB
Delete a track by index. Use live_undo to recover.
Inputs: index (required), confirm. Returns: object
live_env_infoA
Bridge environment info: storage/temp dirs and Live's UI language.
Inputs: no named parameters. Returns: object
live_get_selectionA
The currently selected track and scene in Live's view.
Inputs: no named parameters. Returns: object
live_get_tempoA
Current Live tempo in BPM.
Inputs: no named parameters. Returns: object
live_import_audioC
Import an audio file into the Live project and create an audio clip. If scene_index is given, places the clip in that session slot; otherwise places it on the arrangement at start_time. Requires the .ablx extension bridge.
Inputs: file_path (required), track_index (required), confirm, duration, +3 more. Returns: object
live_is_playingA
Whether Live's transport is currently playing.
Inputs: no named parameters. Returns: object
live_list_tracksA
List all tracks in the running Live set with kind, name, mute/solo/arm state.
Inputs: no named parameters. Returns: object
live_playB
Start Live's transport. Defaults to confirm=True.
Inputs: confirm. Returns: object
live_redoC
Trigger Live's redo.
Inputs: confirm. Returns: object
live_rename_trackC
Rename a track by index.
Inputs: name (required), index (required), confirm. Returns: object
live_render_trackB
Render the pre-effects audio of an arrangement track to a WAV file in the extension's temp directory. Returns the WAV path.
Inputs: end_time (required), start_time (required), track_index (required), confirm. Returns: object
live_set_clip_notesB
Replace the notes in a MIDI clip. notes is a list of dicts with {pitch, start, duration, velocity, mute?}. Existing notes are removed first.
Inputs: notes (required), scene_index (required), track_index (required), confirm. Returns: object
live_set_tempoB
Set Live's tempo (BPM, 20..999). Defaults to confirm=True.
Inputs: bpm (required), confirm. Returns: object
live_statusA
Live transport state: tempo, time signature, playing flag, song position, track count.
Inputs: no named parameters. Returns: object
live_stopB
Stop Live's transport. Defaults to confirm=True.
Inputs: confirm. Returns: object
live_undoB
Trigger Live's undo. Reverts the last LOM-visible change.
Inputs: confirm. Returns: object
live_vocal_to_complement_midiC
End-to-end live flow: render audio from an arrangement audio track range, extract vocal guide, generate complementary melody, create a new arrangement MIDI clip, and write notes into it.
Inputs: end_time (required), start_time (required), target_midi_track_index (required), source_audio_track_index (required), +7 m… Returns: object
midi_estimate_keyC
Estimate the key of a MIDI clip in an .als file using the Krumhansl-Schmuckler algorithm. Returns tonic, mode, and confidence.
Inputs: path (required), clip_index (required), track_index (required). Returns: object
midi_extract_vocal_guideA
Extract a monophonic guide melody from a vocal WAV file with a dependency-free pitch tracker. Returns notes in both seconds and beats (requires tempo_bpm).
Inputs: file_path (required), tempo_bpm (required), hop_ms, frame_ms, +3 more. Returns: object
midi_generate_complementC
Generate a complementary MIDI melody (not a duplicate) from guide notes in beat-space. Input note dicts: {pitch,start,duration,velocity?}.
Inputs: guide_notes (required), seed, density, register, +2 more. Returns: object
midi_guess_chordsB
Guess chords across a MIDI clip with a sliding window. window_beats defaults to 1. Returns time-aligned [{start,end,chord}] segments.
Inputs: path (required), clip_index (required), track_index (required), window_beats. Returns: result
midi_note_statsC
Per-clip note stats: count, range, density, pitch-class histogram.
Inputs: path (required), clip_index (required), track_index (required). Returns: object
midi_vocal_to_complementB
Convenience one-shot: extract vocal guide from WAV and immediately generate a complementary MIDI melody in beat-space.
Inputs: file_path (required), tempo_bpm (required), seed, density, +3 more. Returns: object
project_backupA
Make a timestamped copy of an .als file in /Backup/. Never modifies the source. Returns the backup path.
Inputs: path (required), backup_dir. Returns: object
project_list_versionsC
List timestamped backups for an .als file.
Inputs: path (required), backup_dir. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 40 tools
Tools are generally distinct with clear prefixes for different domains (als, live, midi, etc.). However, there is some overlap between assist_finishing and als_find_unfinished, and between live_vocal_to_complement_midi and midi_vocal_to_complement, which could cause confusion.
Naming follows a consistent pattern with prefixes and verb_noun structure overall. Minor inconsistencies like 'assist_finishing' vs 'assist_harmonic_overview' (gerund vs noun) and the use of 'to' in tool names are acceptable but not uniform.
With 40 tools, this server is on the heavy side. While the coverage is broad across multiple subdomains, it could benefit from splitting into smaller servers or reducing redundancy. The count is borderline high.
The tool surface covers a wide range of Ableton Live interactions: file analysis, live control, MIDI processing, and project management. Minor gaps exist, such as lack of effect/automation tools or session view management, but core workflows are well-supported.
Maintenance
Related MCP Connectors
An MCP server that provides read access to your cloud storage providers, bank accounts and more.
MCP server for progressive tool usage at any scale (see https://klavis.ai)
MCP server for Producer/Riffusion AI music generation
Remote MCP server for audio stem separation, denoise, reverb, and MIDI export.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceMCP server that enables AI agents to read and edit Ableton Live sets, including tracks, clips, MIDI notes, devices, and scenes.1-
- AlicenseAqualityCmaintenanceMCP server that bridges Ableton Live with AI models, enabling real-time project inspection and control such as track overview, device parameters, and audio analysis.1235 PyPI1MIT
- FlicenseBqualityDmaintenanceMCP server for controlling Ableton Live, enabling AI assistants to interact with Live sessions through tools for track/clip/scene management, playback control, and device parameter adjustments.48-
- AlicenseAqualityBmaintenanceAn MCP server that enables AI assistants to control Ableton Live, providing 104 tools for music production including track, clip, device, and mixer control.156867 PyPI73MIT