@mcp-interactive-ui/core

Registry, validators, normalizer, and OpenAI tool adapters. Framework-agnostic; safe to import from any Node ≥ 18 backend.

Install

pnpm add @mcp-interactive-ui/core @mcp-interactive-ui/types zod

Quick start

import OpenAI from 'openai';
import {
  getDefaultRegistry,
  registryToOpenAITools,
  parseOpenAIToolCalls,
  buildSystemPrompt,
  normalizeLLMResponse,
} from '@mcp-interactive-ui/core';

const registry = getDefaultRegistry();
const openai = new OpenAI();

const completion = await openai.chat.completions.create({
  model: 'gpt-4o-mini',
  messages: [
    { role: 'system', content: buildSystemPrompt(registry) },
    { role: 'user', content: 'Show me hot leads' },
  ],
  tools: registryToOpenAITools(registry),
  tool_choice: 'auto',
});

const message = completion.choices[0]!.message;
const rawBlocks = parseOpenAIToolCalls(message);

const response = normalizeLLMResponse(
  {
    text: message.content ?? undefined,
    blocks: rawBlocks,
    messageId: completion.id,
    metadata: { model: completion.model },
  },
  registry,
);

// Send `response` to your frontend. Render it with <RenderAIContent />.

Extending the registry

import { registerBlock, buildRegistry } from '@mcp-interactive-ui/core';

// Phase 1: the closed set is the only allowed types. Custom block types are a
// Phase 2 feature; for now, use `registerBlock` only to OVERRIDE a built-in's
// description/example for your own prompt tuning.

Guarantees

• normalizeLLMResponse never throws.
• Every block it returns passes normalizedBlockSchema from @mcp-interactive-ui/types.
• Unknown block types become a notice with variant: 'warning'.
• Invalid schemas become a notice with the first issue's message.