ableton-mcp-loopback
A loopback-bound MCP server that enables MCP clients to control Ableton Live (set tempo, create tracks/clips, add MIDI notes, start/stop playback) over a secure local-only connection.
README
ableton-mcp-loopback
A loopback-bound Model Context Protocol (MCP) server for Ableton Live. It lets an MCP client (Claude, an agent, an LLM tool runner) drive Ableton Live — set tempo, create tracks and clips, add MIDI notes, start/stop playback — over the standard MCP stdio transport.
This is a security-fix fork of
ahujasid/ableton-mcp (MIT). The
one material change is the network bind: the upstream Remote Script binds its
control socket to 0.0.0.0 (every network interface), exposing unauthenticated
Ableton control to the LAN. This fork binds 127.0.0.1 (loopback) as a hard
constant — the kernel refuses any non-loopback peer at the socket layer, with
no firewall rule required. See Security below.
Why two parts
Ableton's Live Object Model (LOM) is reachable only from a MIDI Remote Script loaded into Live's embedded Python — there is no out-of-process LOM API. So the project is irreducibly two pieces:
RemoteScript/— a MIDI Remote Script that runs inside Ableton Live. It opens a small loopback TCP socket (127.0.0.1:9877) and translates incoming JSON commands into LOM calls (mutating calls run on Live's main thread).ableton_mcp_loopback/— the stdio MCP server. It runs as a normal process, connects to127.0.0.1:9877as a client, and relays MCP tool calls to the Remote Script.
MCP client <—stdio—> ableton-mcp-loopback server <—TCP 127.0.0.1:9877—> Remote Script (in Live) —> LOM
Setup
1. Install the MCP server
With uv (recommended — no manual venv):
uvx --from git+https://github.com/applicate2628/ableton-mcp-loopback ableton-mcp-loopback
Or with pip:
pip install git+https://github.com/applicate2628/ableton-mcp-loopback
ableton-mcp-loopback
Then point your MCP client at the ableton-mcp-loopback command (stdio). For
example, in an MCP client config:
{
"mcpServers": {
"ableton": {
"command": "uvx",
"args": ["--from", "git+https://github.com/applicate2628/ableton-mcp-loopback", "ableton-mcp-loopback"]
}
}
}
2. Install the Remote Script into Ableton Live
This is the one irreducible manual Ableton step — the LOM is only reachable from a Control Surface script loaded by Live itself.
-
Copy the
RemoteScript/folder into Ableton's MIDI Remote Scripts directory, renaming it to a clear name such asableton_mcp_loopback:- Windows:
C:\ProgramData\Ableton\Live <version>\Resources\MIDI Remote Scripts\or\Users\<you>\Documents\Ableton\User Library\Remote Scripts\ - macOS:
/Applications/Ableton Live <version>.app/Contents/App-Resources/MIDI Remote Scripts/or~/Music/Ableton/User Library/Remote Scripts/
The result should be a folder like
.../MIDI Remote Scripts/ableton_mcp_loopback/containing__init__.py. - Windows:
-
Start (or restart) Ableton Live.
-
Open Preferences → Link / Tempo / MIDI (the MIDI tab), and under Control Surface select
ableton_mcp_loopback(the folder name you used). Leave Input/Output set to None. Live then loads the script and you should seeableton-mcp-loopback: Listening on 127.0.0.1:9877flash in Live's status bar.
That's it — the MCP server will connect to the Remote Script automatically. If the server starts before the Control Surface is loaded, it retries the connection.
Tools
This release ships the full upstream ahujasid/ableton-mcp tool surface — it
is a drop-in replacement minus the upstream wide-bind network exposure and minus
telemetry. All 21 tools are present:
| Tool | What it does |
|---|---|
get_session_info |
Tempo, time signature, track counts, transport state |
get_track_info |
A track's clip slots, devices, mixer state |
set_tempo |
Set session BPM |
start_playback |
Start the session transport |
stop_playback |
Stop the session transport |
create_midi_track |
Add a MIDI track |
set_track_name |
Rename a track |
create_clip |
Create an empty MIDI clip in a slot |
create_audio_clip |
Import an audio file into an audio clip slot (Live 12.0.5+) |
add_notes_to_clip |
Write MIDI notes into a clip |
set_clip_name |
Rename a clip |
fire_clip |
Launch a clip |
stop_clip |
Stop a clip |
load_instrument_or_effect |
Load a device onto a track by browser URI |
load_drum_kit |
Load a drum rack and a drum kit into it |
get_browser_tree |
List the browser's category tree |
get_browser_items_at_path |
List browser items at a category path |
switch_to_arrangement_view |
Switch Live's window to the Arrangement view |
set_arrangement_time |
Move the arrangement playhead (beats) |
get_arrangement_clips |
List a track's Arrangement-timeline clips |
duplicate_to_arrangement |
Copy a Session clip into the Arrangement |
A typical end-to-end flow: create_midi_track → set_track_name →
create_clip → add_notes_to_clip → fire_clip.
Input validation and message framing
Every command's parameters are validated before any Live Object Model call
runs (indices are non-negative integers, tempo is within Live's range, MIDI
pitch/velocity are 0–127, durations are positive and finite, note lists are
well-formed). An invalid value returns a structured status: error response and
never reaches Live.
The TCP link between the server and the Remote Script uses newline-delimited
JSON framing: each message is one JSON object terminated by a single \n. Both
ends buffer with a bounded receive buffer, parse each complete frame, reject a
malformed frame without poisoning the stream, and handle multiple frames in one
packet. The wire protocol changed in this release — if you are upgrading from
the earlier P1 build, you must reload the Remote Script in Live (re-select the
Control Surface, or restart Live) so both ends speak the same framing.
Security
Loopback by construction, not by firewall convention.
The Remote Script binds its socket with HOST = "127.0.0.1" as a hard
constant. There is no operator override, no bind-address config key, and no
environment variable that can widen it. Because the socket is bound to the
loopback interface, the operating-system kernel never associates it with the
host's LAN/routable interface, so a connection from any non-loopback address is
refused at the socket layer — you do not need a firewall rule for this, and
no firewall misconfiguration can undo it.
Contrast with upstream ahujasid/ableton-mcp, which binds 0.0.0.0 — that
accepts connections from any interface, meaning any device on the same LAN can
issue unauthenticated commands to Ableton. That is the exposure this fork
removes.
This build also contains no telemetry — the upstream telemetry module, the
user_prompt capture parameter, and the analytics dependency are all removed
(not merely disabled).
Residual risk (local processes). Any process running on the same machine
can still reach 127.0.0.1:9877. This is the same trust posture as every local
stdio MCP server and every local development service. The loopback bind protects
against remote/LAN attackers, not against other local processes on a machine you
already trust. Finer per-tool consent is out of scope for this component.
License
MIT. See LICENSE. Forked from ahujasid/ableton-mcp (MIT,
copyright (c) 2025 Siddharth Ahuja); the upstream copyright is preserved in
LICENSE.
Development
pip install -e .[test]
pytest # security probe + tool-surface tests (no Ableton needed)
python -m ableton_mcp_loopback.server # run the server (will retry to reach Live)
The tests under tests/ are all runnable without Ableton: they assert the
loopback bind constant, scan the tree for any wide-bind literal, exercise a real
loopback socket to prove a non-loopback connect is refused (plus an always-run
structural guard that the bind is loopback, so the security check can never go
green on a skip), check full upstream tool-surface parity on both the server and
the Remote Script, exercise the newline-delimited message framing (partial /
multi-frame / malformed / oversized handling), and exercise the parameter
validation. The full LOM smoke (create track → notes → fire → audible) requires a
live Ableton Live and is run separately.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.