Skip to main content
Glama
dwrightii

WING Rack MCP

by dwrightii

WING Rack MCP

An MCP server that lets Claude control a Behringer WING over OSC on your local network. It talks to the WING Rack, the WING Compact and the full WING. It is for people who build their own scenes: read any node, write named parameters, dump a subtree to a text file, edit it, and push it back.

Version 0.3.1 is the first public release.

What you need

A WING on the same LAN as your computer. Claude Desktop. Python 3.12 or newer on PATH.

There is no credential to enter: the desk speaks unauthenticated UDP on the LAN. Anything on your network that can reach the desk can control it.

Windows is the tested platform. On macOS, python may be absent; if it is, edit mcp_config.command in manifest.json inside the bundle to python3 before you install. Reports from macOS are welcome.

Related MCP server: X32/M32 MCP Server

Install in five steps

  1. Confirm your interpreter and install the one dependency. Run python --version and check it reads 3.12 or newer, then python -m pip install "mcp<2". Claude Desktop launches the server with whatever python is on your PATH, so the package has to be in that interpreter.

  2. Download wing-rack-mcp-v0.3.1.mcpb from Releases and install it in Claude Desktop, under Settings, then Connectors, then install from file. The connector appears as WING Rack (v0.3.1).

  3. Put the computer on the same LAN as the desk.

  4. Ask Claude to run wing_list_consoles. It prints every WING answering on the network. Note the name of yours.

  5. Ask Claude to run wing_connect with that name, then wing_get on /ch/1. A channel of values coming back means you are done.

Tools

Tool

What it does

verify_connection

Health check. Returns the server version, the loaded profile's firmware and whether a desk is connected.

wing_list_consoles

Every WING answering on the network, with its name, address, model and firmware.

wing_connect

Connects by console name, never by address. On a shared network an address match can land on another desk.

wing_get

Reads one or more nodes: each parameter's value, read-only flag and legal range, plus the node's children.

wing_set

Writes named parameters to one node as a single message. Validation is whole-message, so one bad key sends nothing. dry_run shows the payload without sending.

wing_dump

Reads a subtree to a show file, walking child by child, because the desk cannot return a family root such as /ch in one reply.

wing_push_file

Pushes a show file as a background job and returns a job id straight away. Retries the whole push up to three attempts.

wing_push_status

Polls a push job: state, attempt, nodes done against total, elapsed seconds, failure, partial and incomplete counts, and the report file's path.

wing_generate_profile

Re-walks your own desk and rebuilds the command profile. Read-only against the desk.

The shipped command profile was generated on firmware 3.1-0. Regenerate it with wing_generate_profile after any firmware change, so the types and ranges the server validates against match your desk.

Fifteen destructive addresses are excluded from the writable set and unreachable from any tool: card format, recording deletes, scene recall, transport and console-wide save.

The Android companion app

The WING has one analyzer, and whichever desk page is drawing an RTA claims it. wing-rta-toggle is a small Android app that switches the desk's own RTA overlays off so a remote mixing app on a phone can hold the analyzer, and switches them back when you are done. The APK is on its Releases page.

Docs

Safety

The server never deletes from the console's library. A write is read back and compared, because the desk's OK acknowledges receipt and not storage. Connect by console name, never by address. Full rules in docs/safety.md.

Tested on

One desk, one platform: a WING Rack on firmware 3.1-0, driven from Windows. If you run it against a different WING, a different firmware or macOS, a first-time report on Issues is welcome.

MIT licensed. Built by Doug Wright.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    B
    maintenance
    Connects AI coding agents to PreSonus StudioLive III mixers over the local network, exposing live mixer context such as channel names, mute/solo/fader state, Fat Channel compressor/EQ models, and meter activity as MCP resources and tools for reading and reasoning about live sound engineering.
    1
    MIT