Scrape

Command bar

Paste a URL — a bare host like example.com is accepted and upgraded to https:// — and press Enter or Scrape. While the run is live the button becomes Cancel.

The Scrape · Crawl · Watch switch left of the field carries the URL to the other two screens. Copy as… prints the run as CLI, MCP or Go. The cost hint beside the buttons prices any LLM calls before you spend anything, and says when the selector cache makes them free.

Render

Mode	What it does
Auto (default)	Static fetch first; escalates to a real browser only when the page needs JavaScript
Browser	Always render in a browser
Static	Never launch a browser — you want the raw server response

The browser is an installed Chrome, Chromium or Edge; with none, Magpie offers a one-time pinned Chromium download (~150 MB).

Extractor chips

One extractor per run: off, auto (dispatch by URL pattern), schema, or a vertical. The verticals are zero-LLM — GitHub repos, Hacker News, YouTube, arXiv, PyPI, npm, crates.io, Docker Hub, Hugging Face, Stack Overflow, Substack, Dev.to, Reddit, RSS, Trustpilot, Upwork jobs, Amazon, eBay, Etsy, Shopify, WooCommerce and generic e-commerce products, plus article, event, job-posting and local-business pages. Twenty-seven in all; hover a chip for its patterns.

schema opens the panel: describe what you want (“book title, price, stock availability and UPC”), press Draft schema, and an LLM drafts it fitted to the page you scraped — the badge names the provider, tokens and cost. Edit the JSON or YAML, then Save as a name; saved schemas are plain files in the schemas folder, and appear in the Crawl screen’s picker. The full authoring guide is on Schemas.

Advanced

The drawer carries the full option surface. Everything empties to a sensible default; Settings → Network can prefill proxy, fingerprint and language.

Option	Default	What it does
Include (CSS)	empty	One selector per line kept in the cleaned output
Exclude (CSS)	empty	One selector per line dropped — wins over include
Bypass selector cache	off	Skips the learned selectors for this run
Capture XHR	empty	Comma-separated regexps; records matching background responses (Browser render only)
Actions	empty	One browser action per line: click, type, scroll, wait, wait-for, eval-js
Language	Settings default	Accept-Language override
TLS fingerprint	Settings default	Browser TLS handshake: chrome, firefox, safari, edge, ios, chrome_android
Header profile	default	Header bundle: default, chrome, firefox, safari, edge, ios, chrome_android
Cookies	empty	Sent with the request, a=b; c=d
Viewport	empty	Screenshot size, WxH
CDP endpoint	empty	Drive your already-running browser, ws://127.0.0.1:9222
Proxy	Settings default	Per-run egress, http://user:pass@host:port or a pool line; invalid values fail before any request
Provider override	Settings default	LLM provider for this run
Model override	empty	LLM model for this run
API key (this run)	empty	Applies to this run only, never stored
Max cost USD	0 = off	Hard ceiling on this run’s LLM spend

The result workstation

Tabs: Markdown Render, one per picked format, Extracted JSON when a record exists, XHR (URL, status, MIME, body per captured call), Brand, and Viewport. On a format tab, Re-extract reruns the schema over the tab’s content without refetching; with a vertical picked it is disabled — verticals fetch the URL themselves.

Tools: Summarize (1–20 sentences, no refetch), Brand (palette, fonts, logo, favicon — zero LLM), Wrap, Copy and Save as file (the visible tab, under its own name), Discard result.

Error paths and their exact wording are covered in First scrape. For pages behind a login, capture a session first — see Sessions.