Skip to content

AI agents (MCP)

Give Claude Code, Cursor and other agents the rules, so they write code that works the first time.

The @rdloom/mcp server serves the specs: props, when to use each component and when not to, anti-patterns, accessibility requirements, tokens and tested examples. It's read-only and needs no network access.

Set up

npm install -D @rdloom/mcp

Claude Code:

claude mcp add rdloom -- npx rdloom-mcp

Other clients, in their MCP config (for example .mcp.json):

{
  "mcpServers": {
    "rdloom": { "command": "npx", "args": ["rdloom-mcp"] }
  }
}

Tools

  • list_components: all components, or ranked by a search like "pick a date range"
  • get_component: install and import lines, props, usage rules, accessibility, the first example
  • get_example: working example code, each one tested with axe
  • get_component_source: the files add copies
  • get_tokens: CSS variables with light and dark values
  • validate_props: checks props against the spec before the agent writes them
  • get_setup: how to set up a project