@aimata-dsl/mcp-server

Model Context Protocol (MCP) HTTP server for the Aimata DSL compiler.

Overview

@aimata-dsl/mcp-server exposes the @aimata-dsl/core compiler as a local HTTP server that AI assistants (Claude, GPT-4, Gemini, etc.) can call as an MCP tool. This lets AI models compress and expand UI components without bundling the compiler themselves.

Installation & Start

npm install @aimata-dsl/mcp-server
node node_modules/@aimata-dsl/mcp-server/src/index.js

# Or clone the monorepo:
git clone https://github.com/pravesh98/ai-mata-dsl.git
cd ai-mata-dsl
npm install
node packages/mcp-server/src/index.js

Default port: 3100 (override with PORT=8080 node src/index.js)

Endpoints

GET /health

{ "status": "ok", "version": "0.1.0", "uptime": 3.2 }

GET /manifest

Returns the full MCP tool manifest JSON for AI assistant discovery.

POST /compress

Compress JSX/HTML/Tailwind to Aimata micro-DSL.

Request:

{ "code": "<button className=\"px-4 py-2 bg-blue-500\" onClick={fn}>Submit</button>" }

Response:

{
  "dsl": "[button.px-4.py-2.bg-blue-500{onClick=fn}|Submit]",
  "stats": {
    "inputLength": 72,
    "outputLength": 42,
    "compressionRatio": 1.71,
    "inputTokens": 8,
    "outputTokens": 5,
    "tokenReduction": "37%"
  }
}

POST /expand

Expand micro-DSL back to source code.

Request:

{
  "dsl": "[button.px-4.py-2.bg-blue-500{onClick=fn}|Submit]",
  "target": "react"
}

target options: react · html · vue · angular · svelte · solid

Response:

{ "code": "<button className=\"px-4 py-2 bg-blue-500\" onClick={fn}>Submit</button>" }

MCP Tool Integration (Claude Desktop example)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "aimata-dsl": {
      "command": "node",
      "args": ["/path/to/packages/mcp-server/src/index.js"]
    }
  }
}

License

MIT © Pravesh Shukla