Summary: On November 3, 2026, Firecrawl published a comparison of five product data APIs for AI shopping agents: Bright Data, Apify, Oxylabs, Zyte, and Firecrawl itself. The piece frames a concrete builder decision — agents need today’s prices and stock, and a model’s training data has none of it. If you’re wiring a shopping agent, the comparison is mostly about coverage model and per-lookup cost.
Why agents can’t skip the live lookup
A product data API returns listings as structured fields — price, sale price, variants, stock, images, reviews — instead of raw pages. Without one, an agent answers from stale training data. The Firecrawl post cites a Gartner survey published May 27, 2026 in which 54% of consumers who used AI for a purchase said they had to double-check everything the model told them.
The demand side is real too. Adobe figures reported by Digital Commerce 360 show AI-referral traffic to US retail sites grew 62% year over year in July 2026, and those visits converted 60% better than non-AI traffic. Agents are already shopping; the question is whether they answer from live data.
Data reaches an agent two ways: retailer connectors that search a store’s catalog, and page extraction that reads any product URL. Most agents need both — connectors cover the big retailers, extraction covers the long tail of brand stores no connector lists.
Four criteria worth weighing
The post suggests judging an API on four things. Coverage: which retailers have ready-made connectors, and what happens when one doesn’t. Freshness: a live call returns today’s price, while a dataset is a snapshot better suited to analysis. Variant detail: price and stock should sit on each size or color, because that’s how people actually buy. And agent access plus per-lookup pricing — agents make many calls per conversation, so the cost of one lookup multiplied by lookups per chat is your real bill.
How the five options differ
All five expose an MCP server, so the real differences are in how they source data and charge:
- Firecrawl — retailer providers through Firecrawl Alexandria (Shopify, Walmart, Best Buy, Newegg and more), plus a product format on /scrape for any product page. A product page costs 1 credit; most retail Alexandria tools cost 5, with 1,000 free credits a month.
- Bright Data — per-site scrapers and prebuilt datasets, billed per 1,000 records; datasets start at $250 per 100,000 records.
- Apify — community and first-party Actors, billed in compute units on monthly plans from $19; the E-commerce Scraping Tool lists from $1 per 1,000 product details.
- Oxylabs — dedicated scrapers by target (like /amazon/product), from $0.50 per 1,000 results.
- Zyte — AI extraction priced from $0.06 per 1,000 successful responses, rising with site difficulty and browser rendering.
This structure mirrors the approach in my earlier writeup on Alexandria turning source selection into an API call — one account, many providers, cost surfaced per call.
Firecrawl’s angle: one account, both paths
Firecrawl’s pitch is that the same account covers catalog search and arbitrary page extraction. Alexandria tools handle search across retailers, single-store catalogs, reviews, and even barcode matching via Affiliate.com or resale checks on StockX. The /scrape product format reads structured data a store already publishes — JSON-LD, schema.org markup, embedded app state — and merges it without an LLM call. When a page is ambiguous, it returns no product and a warning rather than a guess. /search finds product pages when you only have a name, and /monitor watches pages for price changes.
As a practical starting point: install the Firecrawl plugin in ChatGPT or the Claude connector, sign in, and run a prompt like “find these headphones at Best Buy, Walmart and Newegg — give me price, stock, and which is cheapest today.” You don’t name providers; the assistant picks the tool. Then price out your expected lookups per conversation against the per-call costs above — that math, not the feature list, is what decides it.