Skip to main content
← Lab

Human-AI music lab / Ableton MCP / 2026

Claude Could Write MIDI. It Could Not Read It.

On August 13, 2026, I used Claude Code through a fork of Ableton MCP, a connector that lets an AI model control Ableton Live, to work on a 184-bar minimal techno arrangement in G minor at 134 BPM. Claude could change the set, but it could not read enough back to understand what it had already done.

Claude Code produced the MIDI edits, MCP changes, and supporting tooling under my direction and review. Clarence, my AI system, supplied a skill file with production context. I made the listening decisions and rejected the bland output.

Numbers from the run

22
tools in the original
52
working tools in my fork
28
generated snare notes across 8 bars
23
arrangement snare clips rebuilt
8
phased pad cycles
0
audio paths back to Claude

The limit was read access

The original (upstream) Ableton MCP exposed 22 tools. My fork exposed 52 working tools out of 55 defined commands, including reads for notes, arrangement clips, devices, parameters, automation, groove, mixer state, and probability, plus undo, redo, and track and device management.

Before the fork, Claude could not read enough of the set back to know what the kick was doing before proposing a hat pattern, or see the velocity it was about to adjust.

Capability comparison55 commands defined
With read access, Claude could see what it had done.

Randomness is not groove

The first attempt at making the drums feel alive used random velocity and random note probability. Both were bland, and I rejected them.

Groove is a relationship between events. It requires knowing what came before and what the next hit is preparing.

Timing became grammar

The replacement used explicit timing rules. Anchor hits stayed on the grid. Ghost hits used a quintuplet offset at beat +0.600 followed by a septuplet offset at beat +0.571. The second ghost pulled slightly ahead of the expectation established by the first. Fills used sextuplet timing, which marked a departure without sounding like a mistake.

A generator check on the call-and-response snare grammar produced 28 notes across 8 bars. The snare was rebuilt as two voices across 23 arrangement clips. The kit kept its groove template while the snare remained deliberately ungrooved, so its written timing did not fight a second template.

Call-and-response snare grammar8 bars / 28 notes
Each offset is a named timing relationship.

Phrase structure became visible

A pad used 12-bar cycles against 8-bar material. Those phrase lengths realign every 24 bars, producing slow movement without an obvious reset.

Read access let Claude reason about arrangement form instead of isolated bars.

Phrase phaserealignment every 24 bars
Eight phased 12-bar pad cycles covered bars 57 through 152.

Everything went into one repository

The MCP fork, skill file, timing grammars, pattern files, deploy script, and documentation were versioned together. Pattern files preserved examples from outside sources. Grammars encoded reusable production rules.

A deployment check passed for both Ableton remote-script targets, the local MCP server, and the Claude Code skill. The next session can rebuild the project from the repository.

Unresolved as of August 13, 2026

  • The bass octave may be wrong and still needs a listening decision.
  • Session and Arrangement drums diverged; the correct source is unresolved.
  • Pad presence in the drops and at the peak needs listening.
  • One arrangement automation read path reported false despite an envelope being present.

This run had no audio-return path, so Claude could not hear the result. I could. I have not yet judged how the final arrangement sounds.

What this changes

The useful changes came from read access and explicit musical rules.

I will keep building the read side before adding more write operations, and keep recording musical rules as testable grammars. The open question is whether an audio-return path lets Claude connect those structural decisions to what comes out of the speakers.