Argus for web scraping

Scrapers

29 collectors that open no window at all

The fastest collection is the one that never starts a browser. For the services Argus already has a card for — Maps listings, Instagram and TikTok and X and LinkedIn profiles and posts, Shopify catalogues, four job boards — a scraper takes a form and returns one typed table. The requests go out over plain HTTP through the profile's proxy, carrying a real browser's TLS fingerprint and header set rather than a Python client's. A Maps search that took eight to ten minutes in a window takes about six seconds in none.

Each card is written against one service's own replies and measured against them, so its page says exactly what it asks for and every column of the table it writes — 29 of them, and the column lists are generated from the app rather than described. They cost none of your plan's automation slots, they run from the tab, from a schedule, from an agent over MCP, or as one step in the middle of a bigger workflow. All but 2 of them collect signed out.

What none of them will do is drive a page, and there is no “scrape this URL for me”. When the data is only in rendered HTML, or the job operates an account of your own, the next section is the answer instead.

The browser

And a real browser for everything else

A headless client fails on the things a browser does without being asked: the page's scripts run, fonts and canvas resolve, the layout settles, and the content that only appears after all of that appears. Argus sessions are ordinary Chromium sessions, so the page you collect is the page a person would have seen.

The identity underneath is generated rather than hidden. Each profile gets a coherent hardware identity — platform, CPU cores, memory, screen, timezone, languages, and canvas, WebGL and audio handled in the renderer rather than by a script wrapper. A page reads the spoofed value because it is the only value present, not because something intercepted the real one. That distinction is the whole reason Argus ships its own browser instead of automating someone else's.

Proxies

Your own exits, checked before they are trusted

Argus does not sell you bandwidth and does not route your traffic. You bring proxies from whichever provider you already use, they live in a shared library, and each profile is assigned one. HTTP and SOCKS5 both work, including the authenticated SOCKS5 variant plain Chromium cannot speak at all.

Health checks report the egress IP, the country and the latency, and run automatically for new or failing proxies — concurrently, rather than one at a time. A proxy that fails blocks the launch instead of leaking your real address into a collection run, and a slower background sweep loads real sites through each proxy so a row can name which site refused it. Your workspace picks both the sites in that sweep and which of them are strict enough to stop a launch.

Automation

Build the run without writing the run

An automation is an ordered set of steps on a canvas — navigate, wait, read, click, extract, branch, loop — that executes inside a chosen profile with that profile's proxy, cookies and fingerprint. It is the same session you would have opened by hand, minus the hand.

Runs can start from a profile's start page, from the calendar, or from your own code. Every run is recorded, so you can read back what each step actually did rather than guessing why a field came out empty.

Datasets

The rows land in a table, not in a log

A collection run that prints to a console has produced nothing durable. Argus writes into a dataset: a table your workspace owns, with columns you named and typed — text, number, checkbox, date, select, tags, URL, email, phone. It is stored in the workspace rather than on one machine, so a colleague opens the same table you do.

The save rows step writes today's results; the load rows step reads tomorrow's work list back out, which is what turns a one-off scrape into a standing job. Import CSV, JSON, TXT or XLSX to seed a table or append to it, edit in the grid, filter, sort, freeze the first column, and export the whole thing as CSV or JSON when it needs to leave.

Local API · MCP

Or drive the whole thing from your own code

If you would rather orchestrate this yourself, the launcher exposes a local HTTP API: create and change profiles, assign proxies, update a fingerprint, launch a session, run an automation and read back what it did. Keys are scoped to the folders you name, and can be created, listed and revoked.

It is loopback — on your machine, not ours. There is no hosted endpoint and nothing about your collection passes through us. The same surface is exposed over MCP, so an agent can drive it with the same tools you would, and every request from a new client raises an approve-or-deny prompt in the app before it is answered. Neither the API nor the MCP tools sit behind a plan — both work on the free tier.

Acceptable use

Where this stops

We would rather be specific than vague. Automating your own work in your own accounts, and scraping data that is lawfully accessible to you, are ordinary uses of Argus and our acceptable use policy says so in as many words.

What the same policy prohibits is collecting personal data in breach of data protection law, or processing personal data you have no lawful basis to process — along with signing into accounts that are not yours and attacking or overloading a service. Those lines are drawn where the law draws them, not wherever is convenient.