prompt-to-asset

Prompt enhancer + asset pipeline for developers. One brief → a validated bundle (iOS AppIconSet, Android adaptive, PWA, favicon, OG, visionOS) via routed image models. Three modes: inline_svg (zero-key), external_prompt_only (paste into your subscription)

latest

Source

npm

Version: 0.5.1

Version published: 5 months ago

Maintainers: 1

Created: 5 months ago

Source

prompt-to-asset

One brief → a validated, multi-platform asset bundle. Prompt enhancer + asset pipeline for developers. Works in Claude Code, Cursor, Windsurf, Codex, Gemini CLI, Cline, VS Code, and standalone.

# Zero-key, one-liner
npx prompt-to-asset init

# Or install globally
npm i -g prompt-to-asset
p2a doctor

What you get

One brief + optional brand.json → every asset a modern app needs:

iOS — AppIcon.appiconset (14 sizes + 1024 marketing, opaque), iOS 18 dark / tinted variants.
Android — adaptive foreground + background + Android 13 monochrome, every mipmap-*dpi.
PWA — 192, 512, 512-maskable (80 % safe zone) + manifest.webmanifest + <head> snippet.
Favicon — multi-res .ico, icon.svg with prefers-color-scheme: dark, apple-touch-icon.png.
OG / Twitter — 1200 × 630 with real typography via Satori (not rendered inside a diffusion sampler).
Flutter — flutter_launcher_icons.yaml + 1024 master.
visionOS — 3-layer parallax scaffold (README + master PNG; layer separation is manual).
Splash screens — iOS LaunchScreen-2732.png, Android mipmap-*dpi/splash.png, PWA splash-1200.png.

Plus: logos, illustrations, hero banners, sticker / transparent marks, icon packs. Sprite sheets and 9-slice via the CLI for game devs.

Three ways to run it — pick one per call

Mode	Keys required	Best for	How
`inline_svg`	none	logos, favicons, icons, simple app icons	Server returns a tight SVG-authoring brief; your MCP host (Claude / Cursor) writes the `<svg>` inline, then `asset_save_inline_svg` writes it to disk and runs the platform fan-out.
`external_prompt_only`	none	you already pay for ChatGPT / Midjourney / Ideogram / Google AI Studio / Recraft	Server returns the dialect-correct prompt + paste-target URLs. You generate in your subscription, save the file, then call `asset_ingest_external` to run matte → vectorize → validate.
`api` (free)	none	automation, no paid key	Routes to Pollinations (zero signup), Hugging Face Inference (free `HF_TOKEN`), Cloudflare Workers AI (free 10k neurons/day), or Stable Horde (anonymous queue).
`api` (paid)	one of `OPENAI_API_KEY`, `IDEOGRAM_API_KEY`, `RECRAFT_API_KEY`, `BFL_API_KEY`, `GEMINI_API_KEY`, `STABILITY_API_KEY`, `LEONARDO_API_KEY`, `FAL_API_KEY`	full-fidelity, rate-limit-free	Server calls the routed provider → matte → vectorize → export → tier-0 validate → content-addressed `AssetBundle`.

p2a doctor tells you which modes are live in your current environment.

Why this exists

Devs ship apps. Apps need brand assets. Two obstacles:

You're not a designer. Asking Gemini "iOS app icon for a weather app called Halcyon" returns a mediocre square PNG because you don't know each model's dialect, safe zones, text ceilings, or when transparency is actually supported vs. rendered as a grey-and-white checkerboard.
One PNG is not an asset. A real app icon is a 1024² master + the 14-size AppIconSet + Android adaptive layers + PWA maskable + visionOS + Flutter pubspec. Doing that fan-out by hand in Figma or Photoshop after every regeneration is hours of grunt work.

This tool does for dev assets what paid prompt-enhancer tools do for designer prompts — free, scoped to dev-asset surfaces, routed across every model.

Install

# Per project (keeps your toolchain reproducible)
npm i -D prompt-to-asset

# Global (recommended for CLI use)
npm i -g prompt-to-asset

# Zero install (every command works)
npx prompt-to-asset <cmd>

Register as an MCP server

# Claude Code
claude mcp add prompt-to-asset -- p2a

# Cursor / VS Code / Windsurf — edit .cursor/mcp.json / .vscode/mcp.json
#   "prompt-to-asset": { "command": "p2a" }

# Codex / Gemini CLI
#   Already generated in .codex/config.toml / gemini-extension.json

p2a init scaffolds a brand.json tuned to your framework (Next.js, Expo, Flutter, Xcode, Android, Astro, …) and prints the exact registration command for your IDE.

60-second quickstart (zero key)

npx prompt-to-asset doctor      # what's on, what's missing, ranked best → worst
npx prompt-to-asset init        # scaffold brand.json + register MCP server
npx prompt-to-asset pick        # interactive model picker: asset type + constraints → ranked route

Then, in Claude Code / Cursor / Windsurf:

Generate a favicon bundle for my app called Forge. Use brand.json. inline_svg mode.

Or standalone, from a master PNG you already have:

npx prompt-to-asset export path/to/master.png --platforms all --app-name "Forge"
# → writes iOS AppIconSet, Android adaptive, PWA, favicon.ico, OG, visionOS scaffold, Flutter yaml

Security

API keys live in env vars only. Never written to disk, never logged, never echoed in MCP responses. Provider error bodies are passed through a redact() filter that scrubs common key patterns (OpenAI sk-…, Google AIza…, Anthropic sk-ant-…, Bearer …, ?key=…, Replicate r8_…, Hugging Face hf_…).
Path access is allow-listed. image_path / output_dir inputs are resolved through symlinks and rejected if they escape the project cwd + output dir + cache dir. Widen with P2A_ALLOWED_PATHS=/path1:/path2 if your pipeline writes elsewhere.
SVG is XSS-sanitized before any write. <script>, <foreignObject>, on*= handlers, javascript: URIs, external <image href> / <use href>, CSS @import over the network — all rejected unconditionally. The check runs even without SVGO installed.
Cost guardrail. Set P2A_MAX_SPEND_USD_PER_RUN=5.00 to cap any single tool call. Pre-flight cost estimate (best-effort from published pricing) refuses to call if over. Free-tier routes are always $0.
No telemetry. No remote calls unless the routed provider explicitly requires one.