Scrape by describing what you want, not where it lives

What it does Parsera is a Python library that combines browser automation—it requires Playwright—with an LLM to turn web pages into structured JSON. You hand it a URL and a dictionary where keys are your desired fields and values are plain-English descriptions; it returns a list of records. It works as a library, async coroutine, or CLI tool, and offers a Docker option.

The interesting bit The schema is just natural language. Rather than maintaining brittle CSS selectors, you write prompts like "News title" or "Number of points" and let the LLM figure out the mapping. That shifts the maintenance burden from DOM structure to semantic intent.

Key highlights

• Interface is a single dict: keys become JSON fields, values become LLM prompts.
• Offers sync run, async arun, a CLI with optional page scrolling, and a Docker path.
• Supports swapping in custom LLM backends via configuration.
• Can scroll pages before extraction using the --scrolls CLI flag.

Caveats

• The README never explains how the LLM consumes the page—HTML, text, or screenshot—or what the hosted PARSERA_API_KEY costs.
• The CLI requires an OPENAI_API_KEY while the library uses a PARSERA_API_KEY, and the README does not clarify why the keys differ.

Verdict Worth a look for one-off extraction jobs or prototypes where writing selectors feels like overkill. Skip it if you need deterministic, low-latency scraping or if unpredictable token usage is a dealbreaker.