For AI agents

The library was built with both human and machine readers in mind. Three surfaces make the docs first-class for LLMs and coding agents:

Local MCP server (recommended)

Install @nori-ui/mcp and let your agent run it locally over stdio. The package bundles the entire docs corpus, so there's no network dependency and no rate limit.

Claude Desktop / Claude Code

Add to your MCP config (~/.claude.json, claude_desktop_config.json, etc.):

{
  "mcpServers": {
    "nori-ui": {
      "command": "npx",
      "args": ["-y", "@nori-ui/mcp"]
    }
  }
}

Cursor

~/.cursor/mcp.json:

{
  "mcpServers": {
    "nori-ui": {
      "command": "npx",
      "args": ["-y", "@nori-ui/mcp"]
    }
  }
}

Any stdio-compatible agent

The CLI speaks the standard MCP stdio transport. Point your client at npx @nori-ui/mcp (or install globally and use nori-ui-mcp directly).

Tools exposed

• search_components(query) — fuzzy search by name, description, or tag.
• get_component_docs(name) — full docs body for one component.
• get_component_props(name) — typed prop list with descriptions and defaults.
• list_examples(name) — runnable code snippets keyed by component.

HTTP MCP endpoint (fallback)

For browser-based clients that can't spawn a local process — Claude.ai web, ChatGPT custom GPTs, hosted MCP gateways — the same server is reachable over HTTP:

https://nori-ui.com/mcp

Same four tools, same data, same answers. Prefer the local server when you have the choice; the HTTP endpoint is a quiet fallback.

llms.txt

A flat index of every doc page, optimised for token-efficient discovery. The convention follows llmstxt.org.

https://nori-ui.com/llms.txt

llms-full.txt

Every doc page concatenated into a single plain-text file — the entire library in one fetch.

https://nori-ui.com/llms-full.txt

Use this when context is cheap and you want the agent to have the whole reference up front. For long-running sessions, prefer the MCP server (local or HTTP): targeted, structured, and avoids token bloat.