A two-minute browse is enough.
browser-recon is an AI agent for scraping reconnaissance. You browse a target site like a human; the agent watches; it returns a production-grade scraping plan — recommended library, headers, cookies, rate-limits, cost, and a runnable starter script.
Cuts the reconnaissance phase from a multi-day reverse-engineering exercise to a single coffee break.
Every recommendation is grounded in real test requests through real proxies — not in the LLM's priors.
Output isn't a paragraph of guidance. It's working Python you drop into your stack.
Three commands separate you from a working scraper. Each builds on the last; together they replace several days of manual reverse-engineering.
One line in your terminal: pipx install browser-recon. The CLI weighs 128 KB. No proprietary code lives on your machine.
Chrome opens. You click through what matters — a search result, a product detail, a review listing. Two or three minutes is usually enough.
The agent fires test requests against the site, validates the working approach, and returns a report with the recommended library, headers, cookies, rate-limits and starter code.
Ours is a measurement.
Open any "how to scrape X" tutorial and you'll see the same shape: install a library, copy these headers, hope for the best. When it breaks in production — and it usually does — you're left debugging Akamai's challenge cookies by hand.
browser-recon doesn't write tutorials. It does the experiment for you: fires the requests through real proxies, sees which combination the live site actually accepts, and reports the answer. The recommendation is grounded in what worked, not in what the model expected to work.
We can't tell you exactly how the agent reaches its conclusions — that's our IP. What we can tell you is what comes out the other end.
Identifies Cloudflare, Akamai Bot Manager, PerimeterX, DataDome, Imperva — and what each one will do to a naive scraper.
The agent fires HTTP requests against the target through the proxy tiers you'd use in production, then reports which library × proxy combination actually returned the data.
Measured bandwidth × your proxy rate. Not a vague band; a number tied to data the agent saw.
The report ends with a Python script using the recommended library, headers, cookies and timing. Drop it in and it works.
Every scan, every report, kept for the life of your plan.
Every report opens with the recommendation. The rest of the page is the evidence behind it.
"Cookie warmup required before the Reviews GraphQL endpoint will respond. Estimated cost is $0.40–$2.00 per 1,000 requests. Confidence: 0.82."
Credits expire monthly. Reports stay live for the duration of your tier. Re-scan for one credit to refresh.
A small batch in each week — drop your email, or post about us on X and we'll let you in next.