Skip to main content

Experimental Features

Extra Unstable

While uiMatch itself is in 0.x (experimental) and subject to change, the features on this page are even more unstable and experimental.

They may be removed or significantly changed without notice. These are primarily intended for MCP / AI assistant integration experiments. Please avoid relying on them in long-term CI/CD pipelines.

Experimental Commands​

experimental claude-report​

Generate Claude-optimized comparison report.

Work in Progress

This command is currently a proof-of-concept and not fully functional. The --format option does not produce formatted output due to architectural limitations (the underlying runCompare function does not return CompareResult).

For now, use the standard uimatch compare command output instead.

Syntax:

uimatch experimental claude-report --figma <reference> --url <url> [options]

Options:

  • --format=prompt - Output as LLM prompt (default) [Not yet functional]
  • --format=json - Output as structured JSON [Not yet functional]

Example:

# Currently shows standard compare output with a PoC notice
uimatch experimental claude-report \
  --figma current \
  --url http://localhost:3000 \
  --format=json

Experimental Configuration​

Add experimental settings to .uimatchrc.json:

{
  "experimental": {
    "claude": {
      "format": "prompt",
      "includeRawDiffs": false
    },
    "mcp": {
      "enabled": false
    }
  }
}

Configuration Options​

OptionTypeDefaultDescription
experimental.claude.format"prompt" | "json""prompt"Output format for Claude integration
experimental.claude.includeRawDiffsbooleanfalseInclude raw diff data in output
experimental.mcp.enabledbooleanfalseEnable MCP server integration

Experimental TypeScript API​

caution

The experimental API is subject to breaking changes. Use in production at your own risk.

LLM Formatting​

import { experimental } from '@uimatch/cli';

// Claude-specific formatting
const payload = experimental.formatForLLM(result, { preferTokens: true });
const prompt = experimental.generateLLMPrompt(payload);

Figma MCP Client​

Requires MCP server running.

import { experimental } from '@uimatch/cli';

// Initialize MCP client
const mcpClient = new experimental.FigmaMcpClient(config);

// Get current Figma selection
const ref = await mcpClient.getCurrentSelectionRef();

Use Cases​

AI-Assisted Design Review​

Use the claude-report command to generate reports optimized for AI assistant consumption:

# Generate prompt for Claude Code
uimatch experimental claude-report \
  --figma current \
  --url http://localhost:3000 \
  --format=prompt

MCP Integration​

Enable MCP server integration for enhanced Figma workflows:

{
  "experimental": {
    "mcp": {
      "enabled": true
    }
  }
}

Feedback​

Experimental features help us explore new integration patterns. If you have feedback or use cases, please open an issue.