Gridshift includes a built-in MCP Server that allows AI assistants to read project state and control the DAW — including mixing, plugin management, and MIDI composition.
Install for Claude Desktop
Go to Help → Install Claude Extension in Gridshift’s menu bar. This installs the MCP server as an extension in Claude Desktop.
Add to other clients
The MCP server uses a bridge binary that communicates with Gridshift via Unix socket. Add the following to your client’s MCP configuration:
Gridshift must be running for the connection to work.
Session
Transport
Tracks
Sends & Returns
Plugins
Clips & MIDI
Automation
Samples
Project
Writing a whole project
The tools split into two halves. The mixer, plugin, send and ducking tools nudge
a project you already have open and are listening to. The rest — placing clips as
windows into a file, clip_configure, marker_configure, project_configure,
group_create — build a project from nothing. transport_control sits on both
sides: it drives playback while you listen, and it also builds — set_tempo
before anything is placed, set_loop once there is something to loop over.
The piece that makes the second half work is the audio window. A real
arrangement is not one clip per file; it is a long stem cut into sections, each
clip reading the part of the file that belongs where it sits. Setting
source_offset_beats equal to a clip’s start_beat reproduces exactly that
cut, so every section stays in sync with the rest of the song.
A workable order: set the key and scale with project_configure and the tempo
with transport_control; add the tracks and groups; color them; place the
audio; write the MIDI parts and automation; wire the sends; label the sections
with markers; set the loop region. Then read it back
with get_session_overview and include: ["clips"] — that is how the
assistant checks its own work, and how a later session picks up where this one
stopped.
Design principles
Gridshift’s MCP tools follow best practices for LLM integration:
- Workflow-oriented: Tools map to user goals, not internal APIs.
- Compound operations: Batch changes in single round-trips (e.g. configure multiple tracks at once).
- Rich metadata: Plugin parameters include type info and human-readable labels so the AI can set values correctly without trial-and-error.
- Musical context: Project key and scale are included in the session overview so the AI writes in-key MIDI.
Last modified on September 7, 2026