MCP Server

Why MCP?

Traditional docs are text — the agent reads hundreds of tokens to find a component. MCP gives the agent a structured catalog: search by behavior, read typed schemas, and install components as a tool call. The.schema.ts metadata is the contract.

Claude Code

Add this to the Claude Code settings for your project (or the global settings under ~/.claude/).

// .claude/settings.json
{
  "mcpServers": {
    "hex-ui": {
      "command": "npx",
      "args": ["@hex-core/mcp"]
    }
  }
}

Cursor

Cursor reads a .cursor/mcp.json at the project root.

// .cursor/mcp.json
{
  "mcpServers": {
    "hex-ui": {
      "command": "npx",
      "args": ["@hex-core/mcp"]
    }
  }
}

Available tools

• search_components — fuzzy search over name, description, tags, and AI hints
• get_component — full registry item (props, variants, AI hints, examples, source) for a slug
• get_component_schema — schema only, no source — for agents reasoning about an already-installed component
• list_themes — available theme presets
• get_theme — full token set for a theme (css / json / tailwind)
• scaffold_project — init + starter components in one call
• customize_component — generate a themed variant
• list_recipes — catalog of spec-driven blueprints (auth form, settings page, …)
• get_recipe — ordered install steps, union of peer deps, post-install checklist
• resolve_spec — deterministic brief → ranked component + recipe shortlist
• verify_checklist— cross-check installed components against the registry’s internal-dependency graph

See Spec-driven development for how the last four compose.

Example prompts

• “Find a hex-ui component for a confirmation dialog and add it.”
• “Search hex-ui for a data table primitive and wire it to my users list.”
• “What hex-ui components should I use for a settings page?”