# uxloom-dev/uxloom [Health: Active]

**Category:** 📐 Architecture & Design  
**Repository:** https://github.com/uxloom-dev/uxloom  
**GitHub Stars:** 1  
**npm Downloads (last month):** 517  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/uxloom-dev-uxloom

## Description
Agent-native UI/UX design validation: model user journeys as state machines and prove what AI-generated screens are missing (unreachable states, missing empty/loading/error states, WCAG contrast, touch targets, localization overflow) before code exists — then verify the code implements the contract. Deterministic CI gates (uxloom check, uxloom audit), one-command setup (npx uxloom init).

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

```json
"mcpServers": {
  "uxloom": {
    "command": "npx",
    "args": ["-y","uxloom"]
  }
}
```

## Documentation

## What uxloom-dev/uxloom MCP server does

The uxloom-dev/uxloom MCP server turns interface design requirements into structured project data that agents can inspect and validate. Journeys are represented as state machines, screens are nodes, and each screen can declare a contract for the states it must support. This makes the expected behavior explicit before production UI exists.

Validation covers journey completeness and state coverage, including unreachable screens, dead ends, and absent loading, empty, or error states. The critics also check WCAG contrast, touch-target sizing, and text expansion risks caused by localization. A project can record evidence for design decisions, including rationale, rejected alternatives, sources, and confidence. Review cycles are bounded to three rounds.

The same project can be checked against implementation. `audit` supports web and native markers for Swift, Kotlin, Dart, and Java, and can verify the live DOM. It can also compare a Figma or Penpot export with the design contract.

## How uxloom-dev/uxloom MCP server works

The project file is `uxloom.project.json`, stored in the workspace and versioned with the code. An agent can initialize a project, gather a brief, define journeys, register screens, validate the project, address findings, and produce a coverage report. The preview provides themed, editable mocks and exposes evidence and reviewer comments. Pinned comments can be handed back to an agent with the relevant layout block, screen contract, journey references, and current findings.

The validators are deterministic: identical input produces a byte-identical report. Results can be emitted as JSON or SARIF, and GitHub annotations are supported for CI review. Existing findings can be frozen with an update-baseline option so that newly introduced drift is what blocks a build. Thresholds are configurable through `uxloom.config.json`.

## Setup and configuration

Install the published npm package with `npx -y uxloom`. For Claude Code, the documented MCP registration command is:

```bash
claude mcp add uxloom -- npx -y uxloom
```

Codex CLI can be configured with the equivalent `codex mcp add uxloom -- npx -y uxloom` command. The one-command `npx uxloom init` flow creates MCP configuration, an agent skill, and a starter project file. CI can run `npx uxloom check design/uxloom.project.json`; the command exits with status 1 when errors are present.

Additional CLI operations include `preview`, `export`, `audit`, and `diff`. Exports can produce HTML, SVG, or PNG, with a manifest option for round-trip workflows. No environment variables or credentials are specified in the provided material.

## Tools and capabilities

The uxloom-dev/uxloom MCP server supports an agent workflow built around operations such as `project_init`, `brief_start`, `brief_answer`, `journey_define`, `screen_register`, `project_validate`, and `coverage_report`. Its documented capabilities include:

- Modeling journeys and screen state contracts
- Detecting missing states and unreachable or dead-end paths
- Checking contrast, touch targets, and localization expansion
- Previewing and exporting editable design mocks
- Auditing implementation drift across web and native markers
- Returning CI-friendly JSON, SARIF, and GitHub annotations
- Passing reviewer comments and context to an agent

## Limitations and notes

The JourneyGraph format is currently version `0.1` and may change before 1.0. The provided documentation describes compatibility with Claude Code, Codex CLI, and MCP clients generally, but does not name specific desktop clients. The material also does not specify environment variables, hosted operation, or authentication requirements.

_Full upstream README: https://allmcps.com/mcp/uxloom-dev-uxloom/readme_