DEV Community: Dylan Ferrara

Teach your agent to shop: verify any deal against price history in 20 lines

Dylan Ferrara — Fri, 23 Oct 2026 14:23:54 +0000

Shopping agents are useless if they trust the price tag. Here's the 20-line pattern that makes an agent price-aware using Fazy's free no-key API.

The idea: before your agent recommends a product, it calls price history and quotes the dated facts. "Down $40 from $199 on 3 Sep 2026" beats "60% off!" every time.

import requests

def is_good_price(deal_id):
    h = requests.get(
        f"https://fazy.com/api/v1/deals/{deal_id}/price-history",
        headers={"X-Fazy-Agent": "my-agent"}).json()
    return {
        "lowest_recorded": h["lowest_price"],
        "previous_price": h["previous_price"],
        "facts": h["dated_facts"],  # e.g. "Down $40 from $199 on 3 Sep 2026"
    }

def find_deal(query, max_price):
    deals = requests.get(
        "https://fazy.com/api/v1/deals/search",
        params={"q": query, "max_price": max_price, "in_stock": "true"},
        headers={"X-Fazy-Agent": "my-agent"}).json()["data"]
    return deals[0] if deals else None

Wire the dated facts into your agent's system prompt: "Never quote a discount percentage from the merchant. Quote Fazy's dated price facts."

Prefer MCP? Same tools at https://fazy.com/mcp: search_deals, get_deal, get_price_history. No key, 60 req/min.

I built Fazy (fazy.com). The API is free because agents that verify prices make the whole market more honest, and that's the business.

Black Friday deals are mostly not deals. We checked against the year's lowest prices

Dylan Ferrara — Wed, 14 Oct 2026 00:32:54 +0000

Every November the same thing happens: crossed-out prices appear everywhere, and most of them are theater. So we built the page that answers the only question that matters: is this the lowest price this product has actually sold for?

https://fazy.com/black-friday lists products at the lowest price we have ever recorded for them, each with the earlier price it fell from and that price's date. Rebuilt hourly during the season.

Why this works when "was" prices don't: our was-price study (https://fazy.com/blog/was-prices-we-never-saw) found 83% of claimed was-prices never appeared in our tracking. The anchor is the lie; the history is the truth.

For builders: the same data feeds our free no-key API and MCP server (https://fazy.com/developers). Your agent can answer "should I buy this now or wait" from the record instead of the tag.

I'm the founder of Fazy. The numbers are ours; the method is public; check it.

How we measured 83% of was-prices were fake (methodology deep-dive)

Dylan Ferrara — Fri, 09 Oct 2026 14:21:50 +0000

#3 — How we measured 83% of was-prices were fake (methodology deep-dive)

Tags: data, datascience, webdev, tutorial

We published a finding last week: 83% of "was" prices in our catalog were never the price. Here's exactly how we measured it, so you can replicate or critique it.

The dataset: every product in Fazy's catalog showing a reference ("was") price at least 5% above its current price, with the current price read within the previous three days.

The test per product:

Pull the full price history: every recorded observation with timestamps.
Require at least 30 days of tracking and 4+ observations. Anything thinner is excluded.
Check: did the product ever sell within 2% of the claimed was-price? If no, it fails.

Results: 83% failed. Average watch time was eight weeks with 13 observations each, so this isn't a thin sample artifact.

The category split surprised us: wellness 97%, cosmetics 96%, home 91%, footwear 91%, electronics 88%, appliances 75%, apparel 68%. Apparel's relative honesty makes sense: real seasonal markdowns exist there.

The ugliest stat: 43% of products showed the same price at every single check. Not a sale with a fake anchor — just a fake anchor.

Replicate it: our price history is queryable, free, no key: https://fazy.com/developers and MCP at https://fazy.com/mcp. The get_price_history tool returns the dated facts.

I'm the founder of Fazy, so read skeptically. The methodology is the whole point.

83% of "was" prices were never the price. We measured it.

Dylan Ferrara — Fri, 02 Oct 2026 04:13:06 +0000

We run fazy.com, a deals site that records what products actually sold for. Last week we asked our own data a simple question: when a retailer shows a "was" price, was that ever the price?

For 83% of products, the answer was no. Not once in our tracking history did the product sell within 2% of its "was" price.

How we measured: every product in our catalog showing a reference price at least 5% above its current price, price read within the previous three days, tracked for at least 30 days with 4+ observations. Average watch time: eight weeks, 13 observations each.

By category, the worst offenders: wellness 97%, cosmetics 96%, home 91%, footwear 91%, electronics 88%. Apparel was the most honest at 68%.

43% of products showed the same price at every single check. A crossed-out price that never moves is not a discount. It is the presentation.

Full methodology: https://fazy.com/blog/was-prices-we-never-saw

Why I am posting this here: the dataset is queryable. We expose a free, no-key deals API and MCP server with price history: https://fazy.com/developers and https://fazy.com/mcp. Any agent can ask "what did this actually cost over the last 90 days" instead of trusting the tag.

I am the founder of Fazy, so read the numbers skeptically and check the methodology. That is the point of publishing it.

I wired a free no-key deals API into my agent: 7 MCP tools, verified discounts

Dylan Ferrara — Fri, 02 Oct 2026 04:02:28 +0000

I built fazy.com, a price-checked deals site. Last week I needed deal data inside an agent without scraping, so I used our own API. It is free, no key, 60 req/min, and read-only. Here is the whole integration.

The API: https://fazy.com/developers
The MCP server: https://fazy.com/mcp (7 tools: search_deals, get_deal, get_price_history, similar_deals, list_categories, catalog_slice, post_brief)

Why it is different from scraping: every deal carries a verified discount_pct plus the current and reference price, and get_price_history returns the recorded price changes with dates. So the agent can answer "is this actually a good deal" instead of trusting the store's "was" price.

We ran our own catalogue against those "was" prices: 83% of them were never the price while we watched (methodology: https://fazy.com/blog/was-prices-we-never-saw). That is the dataset your agent gets to query.

Minimal example, curl, no key:

curl "https://fazy.com/api/v1/deals/search?q=robot+vacuum"

Response includes discount_pct, price, reference_price, merchant, in_stock, and a url field per deal. For MCP clients, point at https://fazy.com/mcp and the tools show up.

Repo: https://github.com/FazyFer/fazy-mcp

I am the founder, so obviously biased, but the API is free and the data is the interesting part. Happy to answer questions about the price-tracking methodology.