# aesthete-agent [Health: Active]

**Category:** 📐 Architecture & Design  
**Repository:** https://github.com/docsor1212/aesthete-agent  
**GitHub Stars:** 0  
**npm Downloads (last month):** 137922  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/docsor1212-aesthete-agent

## Description
Mechanically validated frontend aesthetics for coding agents: an evidence-ranked palette library, OKLCH harmony generation, WCAG 2.2 + APCA contrast gates, color-blind simulation and AI-slop detection; installable as an agent skill or MCP server via uvx.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "aesthete-agent": {
    "command": "uvx",
    "args": ["aesthete-agent-mcp"]
  }
}
```

## Documentation

## What aesthete-agent does

aesthete-agent turns frontend color selection into a repeatable validation workflow. Its evidence library contains 119 curated palettes with style tags, like tiers, and gate flags. Agents can search that library by a design brief, generate new harmonies in OKLCH, derive semantic color roles, and evaluate palettes against measurable criteria.

The engine focuses on whether a color system can be used in an interface. Its checks include a five-dimension scorecard, WCAG 2.2 contrast boundaries, optional APCA calculations, three color-blindness simulations, AI-slop signature detection, and a gradient ban that treats any gradient as a P0 issue. An aggregate report returns a single PASS or FAIL result across the applicable gates.

## How it works

The MCP package wraps the engine as JSON-RPC tools over stdio. `uvx aesthete-agent-mcp` runs the packaged server without API keys, and the wheel includes the engine data needed for standalone operation. A pure-Python standard-library fallback is available with `--stdio`; FastMCP is optional for the MCP implementation.

Palette work can begin with evidence matching or harmony generation, then continue through contrast analysis, color-blindness checks, scoring, and reporting. The visualizer renders palettes into ten frontend scene types, including mobile, landing-page, dashboard, e-commerce, and editorial layouts. Variant generation supports visual comparison, while the learning profile records accept and reject verdicts and feeds those preferences back into matching, scoring, and linting.

The underlying engine is also available as an agent skill. Its command-line interface includes operations for matching, scoring, contrast matrices, semantic colors, visualization, reports, themes, exports, environment checks, and related palette tasks.

## Setup and configuration

Install the MCP package from PyPI with:

```bash
uvx aesthete-agent-mcp
```

The server requires no API keys and is designed to work offline. It can also be installed through the Smithery CLI:

```bash
npx -y @smithery/cli install docsor1212/aesthete-agent-mcp --client claude
```

For the skill form, compatible agents can install the repository with:

```bash
npx skills add docsor1212/aesthete-agent
```

The core engine uses Python’s standard library. Pillow is optional for image extraction, and FastMCP is optional for the MCP server. The package is usable from Claude Desktop, Cursor, and other MCP clients; the skill installation path names Claude Code, Codex, and Gemini CLI as additional supported agents.

## Tools and capabilities

The MCP interface exposes these capabilities:

- `palette_match` ranks evidence-library palettes against a brief.
- `palette_harmony` generates OKLCH-based harmonies.
- `palette_score` and `palette_report` calculate the scorecard and aggregate gate verdict.
- `palette_contrast` and `palette_matrix` check individual pairs or full WCAG/APCA matrices.
- `palette_blind` simulates color-vision differences, while `palette_semantic` generates categorical colors intended to avoid collisions.
- `palette_slop` detects specified AI-slop signatures.
- `palette_visualize` creates ten-scene renderings and variants.
- `palette_themes` and `palette_theme` work with named theme packs.
- `palette_export` produces CSS, Tailwind, W3C Design Tokens, context, or prompt output.

The broader engine includes 38 subcommands, five full-page layout recipes, eight named theme packs, and export support for CSS, Tailwind, W3C tokens, context, and image-generation prompt blocks.

_Full upstream README: https://allmcps.com/mcp/docsor1212-aesthete-agent/readme_