AI agents

Give your agent the llms.txt index for installation guidance, component contracts, and plain Markdown references. Use llms-full.txt for the complete reference, or agent-context.json for component APIs, generated native examples, token definitions, and package versions.

Local MCP access

From a repository checkout, run pnpm install --frozen-lockfile followed by pnpm agents:build. Configure your MCP client with the absolute path to the server:

{
  "mcpServers": {
    "aidsgn": {
      "command": "node",
      "args": ["/absolute/path/to/aidsgn/packages/mcp/src/cli.mjs"]
    }
  }
}

The server reads a bundled snapshot, works offline after installation, and needs no credentials. It provides component discovery, exact APIs and examples, document search, document reading, and token search. Resources expose the catalog and individual documents. The separate @aidsgn/mcp package is prepared for publication through the normal release process; use the source configuration until its first npm release.

Generate reliable interfaces

Ask the agent to discover components before generating code and to read the selected component’s full contract. Prefer the React and Vue adapters for those frameworks. Other frameworks and server templates can use the canonical native HTML in the catalog while preserving its semantic descendants, IDs, and ARIA relationships. Browser controllers attach behavior to existing light DOM; they do not build the component tree.

Compare the returned package versions with the application’s installed versions. The deployed website follows its repository revision; a local MCP package keeps its bundled snapshot. Token results identify exact names, intended usage, authored aliases, and theme/brand source files.

See the complete agent reference for tool names, framework boundaries, and context maintenance.