AI shopping assistants blend conversational interfaces with real-time product data to help users discover items, compare prices, and check availability. By late 2025 this space sits between generalist LLM-driven research (ChatGPT-style agents using web connectors) and dedicated retail AI startups that build end-to-end extraction, normalization and integration stacks. Key technical building blocks include Web Data Extraction (Playwright, Firecrawl), Tool Integrations and API orchestrators (Pipedream), Chat API Integrations and MCP adapters (Skyvern, Playwright MCP, Chrome DevTools MCP, Exa), Data Pipeline Orchestration (Dagster), and Database Connectors plus Localization/Translation and Text Transformation layers for normalized, multilingual outputs. Trends driving relevance: LLMs are increasingly able to reason about product pages when paired with Model Context Protocol (MCP) browser automation, enabling structured accessibility data rather than brittle pixel scraping. Vendor-agnostic MCPs let different LLMs control browsers for live checks; dedicated scrapers like Firecrawl and search-oriented MCPs like Exa provide complementary discovery capabilities. Orchestration tools (Dagster, Pipedream) chain scraping, enrichment, translation and storage steps into reliable pipelines. This split—conversational research vs specialized retail stacks—matters for accuracy, latency, compliance, and scalability: generalist agents are fast to iterate and good for exploratory queries, while startups focus on robust crawling, normalization, repeatable pricing feeds and localization for commerce use-cases. Understanding these components helps teams choose between quick, LLM-first shopping research flows and production-ready retail AI systems that integrate scrapers, MCPs, pipelines and databases for operational use.