Sound effects and UI sounds
Up to 16 per call from 28 recipes, from coin and laser to click and chime, or from your own notes. UI sets share one character and come with playback hints for cooldown, overlap and gain.
Moose Tracker MCP
Moose Tracker MCP connects your assistant to a full synthesizer, sampler and tracker. Your AI writes songs, game soundtracks and sound effects as notes. Moose Tracker plays them, measures the mix, and hands back finished audio and a project you can open and edit in your browser.
Write a calm exploration theme for my game, two minutes, looping, with a tense layer for when enemies show up.
Why it matters
A language model can describe a song and print a list of notes. It can't play them, can't tell that the bass is drowning the kick, and can't hand you a file. Moose Tracker gives it the missing half: an engine, a pair of measured ears, and a format people can actually work with.
Headline features
Not a clip generator. A complete studio your assistant drives with notes, from the first sketch to the file you ship.
Songs run up to 16 channels, and any channel can split into 16 sub-channels, so piano chords sustain and guitar notes ring over each other the way a player would let them. Each part plays a synth voice or a sampled instrument: pianos, strings, brass, reeds, guitars, organs, voice and full drum kits, with velocity layers so a soft note sounds softly played.
After every edit, a free light analysis returns errors with line numbers, the real length, the key and the energy curve, plus a score from 0 to 100 with up to five specific fixes, judged against a profile for one of 99 genres. A deep analysis renders the song and measures loudness and peaks section by section, then names the part, the bar and the change to make.
The rules are fixed and repeatable. No AI grades the song.
One render turns a song into an adaptive kit: a looping layer per role, each exactly the same length down to the sample, grouped into intensities from calm to full. Your game fades between them as the action builds, and fires one-shot stingers for a win or a boss entrance.
In clients that display MCP Apps, including Claude and ChatGPT, every render arrives as a player: play and pause, a waveform you click to seek, loop, and an A/B switch against the previous version. It lists the sections and instruments, and one button opens the song in Moose Tracker. Other clients get the same links as text.
Try it here: press play on a few demo songs by the Moose.
Also included
Eight tools cover the whole job, from a single button click to a finished soundtrack.
Up to 16 per call from 28 recipes, from coin and laser to click and chime, or from your own notes. UI sets share one character and come with playback hints for cooldown, overlap and gain.
Presets are a starting point. When none fits, your AI designs a new voice, analog, FM or chip, and hears it in about a second.
A curated library covering about 50 styles gives advice with ranges for every stage, from harmony to the final mix. It returns advice only, never notes or audio.
Bring songs in from MIDI and MOD, XM, S3M or IT files, or send them out to a DAW, a notation app or another tracker.
WAV at 16-bit 44.1 kHz or 24-bit 48 kHz, Ogg Vorbis with sample-exact loop points, and Opus. Stems per role, named cues and exact lengths.
Loop renders fold the reverb tail back into the start, so a level theme repeats with no gap and no click.
Every song gets a link that plays it in the browser, with nothing to render. Links are encrypted so we can't read them, and expire after 90 days.
Songs travel as compact text, not audio. A complete looping theme fits in under half a kilobyte, so long sessions stay fast and cheap.
The same notes always sound the same. Any render can be reproduced exactly, and every analysis follows fixed rules.
How it works
A theme, a jingle, a set of UI sounds, a loop for a video. As loose or as specific as you like.
It can research the style first, lay out the sections, then write each part as MTL, Moose Tracker's compact song text.
Errors, length, score and mix flags come back after every edit, and your AI fixes what they find.
Audio files for your project, a link that plays in the browser, and a project you can open and keep editing.
song "Pixel Run" bpm 140 spd 6 ch 6 master 80 loop on
style chip
i1 @lead
i2 @bass
i3 @arp
i4 @drums
P A 32 /2
1 i1 v50: A4 . C5 . E5 . C5 . F4 . A4 . C5 . A4 .
2 i2 v56: bass(Am F x8 root o2)
3 i3 v36: arp(Am F x8 up o4)
d kick: x... x... x... x...
d snare: ....x.......x...
d hat: euclid(8,16)
P B 32 /2 = A
1 i1 v50: C5 . E5 . G5 . E5 . G5 . A5 . G5 . E5 .
order A | A B A BNothing on the server writes or composes music. Your AI writes it; Moose Tracker plays, measures and converts it, and the song stays open for you to change.
Built for
Level themes that loop cleanly, adaptive layers, stingers and a full set of effects, in formats your engine already reads.
Clicks, toggles, confirmations and alerts designed as one family, so every new sound matches the last.
A cue cut to an exact length. Ask for it calmer or without drums and your AI revises the arrangement.
Let your AI sketch chords, bass and drums, then open the song in Moose Tracker and finish it by hand.
Get started
Works in any client that supports MCP, such as Claude, ChatGPT and Grok. No sign-in, no API key.
Moose Tracker MCP is a remote server over Streamable HTTP.
https://mcp.moosetracker.app/mcpAdd a remote MCP server with the URL above. No sign-in or API key.
Tools that start a local command instead can run:
npx -y moosetracker-mcpAdd a custom MCP server with the URL above and “No authentication”.
Add a custom connector with the URL above and “No sign in”.
Or open the .mcpb file in Claude Desktop (needs Node.js 18+).
Run:
claude mcp add --transport http moose-tracker https://mcp.moosetracker.app/mcpOpen MCP Servers, then Configure MCP Servers, and add a remote server with the URL above.
Click Add to Cursor, or add the URL above to ~/.cursor/mcp.json.
On grok.com, add a custom connector with the URL above and no authentication.
In Grok Build, unzip the plugin and run grok plugin install <folder> --trust.
Click Add in VS Code, or add the URL above to .vscode/mcp.json.
Open MCP settings in Cascade and add a server with the URL above.
Try one of these, or describe your own.
Step-by-step guides for each client, plugin packages with checksums and paste-in instructions are on Use Moose Tracker with your AI. After a Moose Tracker update, refresh or reconnect the server in your client, or start a new conversation, to load the latest tools.
Questions
No. Your assistant writes every note. Moose Tracker is the instrument: it plays, measures and converts what your AI writes, and the same notes always sound the same.
Moose Tracker MCP is free to use, with fair-use limits. Your AI can ask the server for the current numbers at any time.
Use what you make anywhere: apps, games, films and ads, commercial work included. No royalties or credit owed for Moose Tracker sounds, presets, demos or the sample library. The Content license has the details.
No. There is no sign-in and no API key. Add the server address to your client and start asking.
Rendered files are kept for about 2 hours, so download what you want to keep. Share links are encrypted and expire after 90 days. Privacy explains the rest.
The MCP reference covers every tool, the analysis flags, file formats, adaptive kits and limits.
Add one server address and your assistant can write, hear, fix and deliver music you can keep working on.
Prefer to write it yourself? Moose Tracker runs in your browser, with no install and no account.