llms.txt
+26 −5
https://moosetracker.app/llms.txt
− # Moose Tracker− > Moose Tracker is a free retro synthesizer and music tracker that runs in the browser (https://moosetracker.app). AI assistants can use its public MCP server to make sound effects, UI sounds and songs as audio files. It is not an AI music model: you write the notes, the Moose Tracker synthesizer plays them.− - Tools (seven): describe, sfx, song, render, convert, assets, feedback. Audio is WAV by default; Ogg Vorbis or Opus on request. Assistants can design custom instruments (audition one voice, start from role templates). Whole pieces can also be composed from musical generators (`compose`: form, chords, bass and drum parts). Call `describe` first (no arguments) for an index and how to write songs.− - Synth or samples, part by part: besides the synth there is a library of CC0 sampled instruments (pianos, basses, strings, brass, reeds, woodwinds, guitars, mallets, organs, voice, jazz drum hits). Use samples for acoustic or real-band parts, the synth for chip, game, UI and electronic sounds, pads, synth basses and leads or anything that needs shaping; blends are often best. MTL: `i2 sample "upright-bass"`. `describe(topic="samples")` lists the packs live with playable ranges (filter with q, c− - [Use Moose Tracker with your AI (setup for Grok, Claude, ChatGPT, Cursor, VS Code)](https://moosetracker.app/ai)+ # Moose Tracker MCP+ > Moose Tracker MCP is the public MCP server for Moose Tracker, a free synthesizer, sampler and music tracker that runs in the browser (https://moosetracker.app). AI assistants use it to make sound effects, UI sounds and songs as audio files. It is not an AI music model: you write the notes, the Moose Tracker synthesizer plays them.+ - Product: Moose Tracker MCP+ - Tools (eight): describe, sfx, research_composition, song, render, convert, assets, feedback. research_composition gives composition advice from a curated corpus covering about 50 styles evenly; no AI composes or generates music on the server, a semantic search index helps find relevant advice. Songs can be built structure first (song plan, one batch of layer edits, a one-section preview) and mixed with the app's five volume layers (velocity, sound gain, channel fader, song level, limiter); ren+ - Synth or samples, part by part: besides the synth there is a library of sampled instruments, CC0 except the TR-808 kit (its author released it as "absolutely free"): pianos, basses, strings, brass, reeds, woodwinds, guitars, mallets, organs, voice, drum hits and drum kits (rock-kit, buzz-kit, roots-kit, tr808-kit; one instrument, each drum on its General MIDI key: C2 kick, D2 snare, F#2 closed hat). Use samples for acoustic or real-band parts, the synth for chip, game, UI and electronic sounds+ ## Outputs: listening and embedding+ - Share link: song (and its analysis) returns a moosetracker.app link that plays the song in the browser for 90 days and needs no render; it suits a person who wants to listen.+ - render: produces WAV, Ogg Vorbis or Opus files to download and embed in another product such as a game, app, presentation or video. For a person who only wants to listen, the share link from song plays the song in the browser and needs no render.+ - sfx: produces sound-effect files to download and embed in an app, game, UI or video; the reply links each file for a quick listen.+ ## Polyphony+ Polyphonic instruments (piano, harp, guitar, pads, chords on any ringing sound): the note-list path is the default. song(layers:{notes:[{ch, ins, notes:[[pitch, startBeat, beats, vel]]}]}) (vel 0-64, the same scale as vNN) sends the notes to the engine's placeNotes, the allocator the piano roll uses: each note goes to a free sub-channel of ch (up to 16), and the reply lists where each note landed plus any warnings (over 16, collisions). Manual lanes (ch 3b) give exact control.+ ## Single-channel audition+ render(mtl or ref, audition:{ch, try?}) plays one channel alone as a short preview on the audition quota; try swaps that channel's sound (an instrument line such as i3 "Warm Pad"), which suits comparing the alternatives light analysis suggests.+ ## Analysis stages+ - Stages: plan, write, light analysis loop, one deep analysis, fix; then the share link to listen, or one render to embed.+ - Light analysis, song(analyze:true): no audio, instant, free. Suited to every edit while writing; the reply ends with a next step (fix the flagged items, or deep analysis when clean).+ - Deep analysis, song(analyze:true, deep:true): an analysis-only render with no audio file. Suited to once or twice when the song is nearly done, and to compare versions; the reply ends with a next step (fix the flagged items, or keep the song as it is (the share link plays it; render produces files for embedding)).+ ## Install+ - Setup page for Moose Tracker MCP, with per-client steps: https://moosetracker.app/ai+ - Generic remote URL: https://mcp.moosetracker.app/mcp (Streamable HTTP, no authentication)+ - Local stdio bridge: `npx -y moosetracker-mcp` (forwards to the live server; holds no engine code and no secrets)+ - Official CLI: `claude mcp add --transport http moose-tracker https://mcp.moosetracker.app/mcp`, and the same form with `gemini mcp add`+ - Packages, with checksums on the setup page: Grok plugin zip, Claude Code plugin zip, ChatGPT/Codex Agent Plugins zip, Claude Desktop `.mcpb` (local stdio bridge)+ - After an update: the server updates in place, but a client keeps the tool list it fetched. ChatGPT: Settings → Apps → Moose Tracker, press Refresh (no reinstall). Codex: `codex plugin marketplace upgrade`. Claude: start a new conversation. Every tool reply carries `serverVersion`; describe's description names the version the tool list came from.+ - Clients that only start a command (Windsurf, Cline, Continue and others): a JSON or YAML config with `command: npx` and `args: ["-y", "moosetracker-mcp"]`. The setup page has the snippets.+ - [Use Moose Tracker MCP with your AI (setup for Grok, Claude, ChatGPT, Cursor, VS Code)](https://moosetracker.app/ai)