| name | generate-music |
| user-invocable | true |
| argument-hint | ["genre / mood / instrumentation"] |
| description | Compose an original music track on TITLES (titles.xyz) from a text description. Use when the user says: "make music", "generate a track", "a beat", "background music for X", "an instrumental", "lofi for a podcast intro", "a jingle", "a song about Y". Describes genre / mood / instrumentation as a prompt; optional lyrics (sung verbatim) or instrumental-only; optional duration. Returns an audio track on the canvas plus a downloadable file. Runs on the TITLES MCP — if TITLES tools are missing, connect mcp.titles.xyz/mcp first (see titles-setup). NOT for: spoken narration or voiceover (text-to-speech), or sound on a video clip (generate-video / animate-image with audio).
|
generate-music
Text description → an original music track.
Get connected
Check the tool list for TITLES tools (names contain titles_). If missing, hand off to the titles-setup skill and stop — never fall back to a non-TITLES tool.
Compose
Get the brief — genre, mood, instrumentation, tempo — described as the sound, not artists or labels (provider moderation hard-fails prompts naming real artists/songs, so translate "sounds like [artist]" into sonic qualities). If they want a particular model for genre fit or lyric support, match one with titles_search_models({ operator: "txt2MusicNode" }) (they differ on genre strengths and whether they sing lyrics) and pass its model_id; otherwise omit model_id and let the server pick a default. Then the one call, titles_generate_music, with that brief as the prompt; words to sing go in lyrics; ask lyrics-vs-instrumental if it's a "song."
You can't hear the result — don't describe how it sounds; hand over the listen link and let the user judge.
The connected server is authoritative for everything else — exact inputs, model resolution, cost approval (price_confirmation_required → max_price_usd), session/canvas handling, sourcing output_ids, and bringing outside images in (titles_create_upload) all follow the tool's own description and the server instructions, not anything memorized here. titles_help has the current catalog.
Deliver the file
Besides the session_url the server points you at, hand over the actual file via titles_download_asset — host-adaptive: on a shell host (Claude Code / Codex) curl it to disk and give the path; on a chat host (claude.ai / mobile) give the short-lived link to click, re-fetching cheaply if it expires.
