Build Your Own Amazon Price Tracker in Python

The whole system is one daily batch job. Nothing runs between batches, so there is no server to babysit.

The watchlist is deliberately boring. Prices are stored in integer cents everywhere, so target_cents is too. A blank target means "track it, never alert":

asin,label,target_cents
B0EXAMPLE1,Espresso grinder,18999
B0EXAMPLE2,Standing desk frame,
B0EXAMPLE3,Replacement filter 3-pack,2450

Install three packages into a virtual environment on Python 3.11 or newer. The floor comes from matplotlib: version 3.11.2, released on PyPI on 2026-09-11, declares Requires-Python >=3.11.

python -m venv .venv
. .venv/bin/activate
pip install "requests>=2.34" "selectolax>=1.0" "matplotlib>=3.11"

If you sell on Amazon and only care about your own listings, the Selling Partner API is the cleaner source, and our breakdown of which proxies Amazon sellers actually need starts from that point. This tracker is for ASINs you don't control: competitors, suppliers, or things you want to buy.

Three decisions matter for this job, and each one has a right answer for a daily ASIN sweep.

Country. Amazon decides currency, default delivery region and sometimes the offer itself from where the request comes from. For amazon.com, exit in the US. SparkProxy's USA Rotating pool is listed on the homepage (read 2026-10-30) as "50,000+ premium US datacenter IPs", so every exit is already American and you don't need a country flag at all. On a Worldwide Rotating plan (1M+ IPs, 80+ countries), the dashboard's credential widget shows a country suffix in the username, displayed on the homepage as user-country-us. Copy the exact string your dashboard generates rather than typing it from memory.

Rotation mode. SparkProxy offers Random Rotation, a new exit IP on every request, on port 11000, and Sticky Sessions, which hold one IP for about five minutes, on port 11002. A product-page fetch is a single stateless GET, so Random Rotation is the right fit. It also gives you free retries: when one IP draws a challenge, the next attempt leaves from a different address. Sticky Sessions only earn their place if you set a delivery ZIP through the location picker and need the follow-up page view to come from the same IP. That flow is explained in our guide to how sticky session proxies hold an IP.

Concurrency. SparkProxy plans are sold by threads, with unlimited bandwidth on every plan. The Starter plan's 100 threads are far more than a tracker needs. A daily job gains nothing from running hot, and a burst of parallel hits is the easiest pattern for any anti-bot system to spot. Six workers is the default below, and you can raise it once your fetch_log shows a low challenge rate.

Be clear about what a proxy can and cannot fix here. This table saves a lot of misdirected debugging:

ZIP pinning has its own write-up in our Amazon product data scraping guide, along with legal framing and search-page pagination. Neither is repeated here.

The fetcher does exactly one thing: make one request through the rotating gateway and return the status and body. It opens a fresh requests.Session per call so no Amazon cookie from a challenged attempt follows the next one.

# fetch.py: one GET per attempt through the rotating US gateway
import os
import requests

USER = os.environ["SPARK_USER"]      # copy both from the SparkProxy dashboard
PASSWORD = os.environ["SPARK_PASS"]
ROTATING = f"http://{USER}:{PASSWORD}@gateway.sparkproxy.io:11000"

HEADERS = {
    "User-Agent": ("Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
                   "(KHTML, like Gecko) Chrome/130.0.0.0 Safari/537.36"),
    "Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8",
    "Accept-Language": "en-US,en;q=0.9",
}


def fetch_via_proxy(asin: str) -> tuple[int, str]:
    """Return (http_status, html). A fresh session per call means no cookie
    carry-over, so each attempt looks like a new visitor on a new IP."""
    with requests.Session() as s:
        r = s.get(
            f"https://www.amazon.com/dp/{asin}",
            headers=HEADERS,
            proxies={"http": ROTATING, "https": ROTATING},
            timeout=30,
        )
        return r.status_code, r.text

Request the bare /dp/ URL. The long, keyword-stuffed URLs you copy from a browser carry ref= and tracking parameters that change nothing about the price and make your request log harder to read. If you whitelist your server's IP in the dashboard instead of using user:pass, drop the credentials from ROTATING and keep the host and port.

Before trusting the gateway, confirm rotation from a shell. Two calls should print two different addresses:

curl -s -x "http://$SPARK_USER:$SPARK_PASS@gateway.sparkproxy.io:11000" https://api.ipify.org; echo
curl -s -x "http://$SPARK_USER:$SPARK_PASS@gateway.sparkproxy.io:11000" https://api.ipify.org; echo

This is the part that quietly ruins most homemade trackers. When Amazon decides a visitor looks automated, it doesn't always send an error. It sends a small page with a 200 status, and a scraper that trusts the status code parses it, finds no price, and writes a NULL or, worse, a zero into your history.

We captured what that page looks like today. On 2026-10-30 we requested two product pages directly from one of our own servers, with no proxy, a desktop Chrome User-Agent and Accept-Language: en-US. Both requests got the same response: HTTP 200, 3,781 bytes, with these features: