prompt-to-asset MCP: Routed Image Generation for Developers

What is prompt-to-asset MCP?

It is a stdio MCP server, packaged with its own CLI, that turns a one-line brief into production assets: logos, app icons, favicons, OG images, illustrations and splash screens. The same 24 tools back both surfaces, so a CI script and a chat assistant hit identical code.

The three generation modes

• inline_svg has the host model write SVG directly, then asset_save_inline_svg validates it and writes the bundle, no provider key needed.
• external_prompt_only returns a ranked, free-first paste target, then asset_ingest_external mattes, vectorizes and validates whatever comes back.
• api calls a routed provider directly, zero-key via Cloudflare Workers AI, NVIDIA NIM, HF Inference or Stable Horde, or a paid key if one is set.

What zero-key first means

asset_enhance_prompt ranks free routes ahead of paid ones on every call: Cloudflare Workers AI, NVIDIA NIM, HF Inference and Stable Horde cover diffusion models, and Pollinations needs no signup at all. A paid key only gets used when a job specifically calls for it.

What validation checks before shipping

asset_validate runs tier-0 checks on every output: pixel dimensions, alpha channel presence, a checkerboard FFT that catches a fake transparent background rendered as pixels, a safe-zone bounding box, ΔE2000 palette drift against the brand, WCAG contrast and OCR Levenshtein distance on any wordmark text.

Who prompt-to-asset MCP is for

Best for

• Developers who want an app icon, favicon or OG image bundle without picking an image model or opening a billing console first.
• CI pipelines that need the same asset generation the chat assistant uses, since the CLI and the MCP tools share one core.
• Teams that want shipped assets checked for a transparent-background failure or contrast problem before a release, not after a bug report.

Not for

• When the input is already a finished SVG that needs Android Vector Drawable XML output instead of a new generated asset, the SVG to Android Vector converter is the better fit for that single conversion.
• When the deliverable is one animated SVG rendered from a JSON config with a PNG preview, rather than a routed production asset bundle, nakkas MCP Server is the better fit for that render.
• When the need is loading general visualization design guidelines for charts or mockups on demand rather than generating a finished asset bundle, Generative UI MCP is the better fit for that guidance.
• When the job is producing mockups, icons and illustrations with prompt patterns, a curation loop and brand compliance scoring rather than routing across image models directly, the AI Design Generation skill is the better fit for that process.

Connect prompt-to-asset MCP: Routed Image Generation for Developers to your AI client

claude mcp add prompt-to-asset -- p2a

1. Register it with Claude Code

   Run claude mcp add prompt-to-asset -- p2a once the package is available on PATH, or use npx directly as the command in the config above.

2. Install the package

   Run npm install -g prompt-to-asset for a global p2a command, or npm install -D prompt-to-asset per project for CI; every command also works through npx with no install step.

3. Or register with another IDE

   Cursor, VS Code, Windsurf, Codex and Gemini CLI each take the same command and args shown above, added through that client's own MCP settings.

4. Check what is live

   Run npx prompt-to-asset doctor, or ask the assistant to call asset_doctor, to see which free routes, paid keys and modes are available right now.

Client configuration

{
  "mcpServers": {
    "prompt-to-asset": {
      "command": "npx",
      "args": ["-y", "prompt-to-asset"]
    }
  }
}

What you can do with it

• Zero-key first routingThe router tries Cloudflare Workers AI, NVIDIA NIM, HF Inference, Stable Horde and Pollinations before any paid model.
• Three generation modesinline_svg, external_prompt_only and the api mode mean a brief can finish with no account at all.
• Validates before shippingasset_validate checks dimensions, alpha presence, a checkerboard FFT, safe-zone bounds, palette drift, contrast and wordmark legibility on every output.
• Research-backed routing rulesEach rule in asset_enhance_prompt's routing trace cites its research source, plus which models get refused for that job.
• Offline platform fan-outasset_export_bundle turns one master image into iOS, Android, PWA, visionOS and Flutter icon sets with no network call.
• Brand bundle awarenessOnce a brand definition exists, every generator reads its palette, fonts and style references automatically.
• Sprite sheets and 9-sliceasset_sprite_sheet and asset_nine_slice pack frames or configs for Phaser, PixiJS, Godot or Unity engines.
• Keys stay in environment variablesProvider keys are read from environment variables only, and provider error text is redacted before reaching the model.

Tools it exposes

Frequently asked questions