WebReaper ships an Agent Skill for Claude Code. Once it is installed in your project, you stop thinking about the CLI flags: you describe what you want in plain language and Claude Code picks the right webreaper invocation for you.

Install the skill

From your project root, run:

webreaper init

This writes a skill manifest to .claude/skills/webreaper/SKILL.md. The file describes what WebReaper can do and which subcommand fits which kind of request, so Claude Code can discover and use it automatically.

How routing works

When you ask Claude Code to do something that looks like scraping, crawling, or extraction, it reads the skill and shells out to the CLI:

"Read this page into Markdown" maps to webreaper scrape <url>.
"What pages does this site have" maps to webreaper map <url>.
"Crawl the whole site" maps to webreaper crawl <url>.
"Pull these fields from the page" maps to webreaper scrape <url> --schema ....

The skill also knows when WebReaper is the better choice than a plain page fetch: when you want clean Markdown to work with downstream, structured fields, multiple pages, or a JS-rendered or bot-protected site.

Example prompts

These all route to WebReaper once the skill is installed:

Scrape https://news.ycombinator.com and give me the front page as Markdown.
 
What URLs are under /blog/ on https://example.com?
 
Crawl https://example.com and save every page to pages.jsonl.
 
Pull the title, price, and tags from this product page.
 
This site is behind Cloudflare and blocking me. Get me the content anyway.

For the bot-protected case, Claude Code can add --browser and --auto-stealth so the CLI renders the page and escalates to a stealth browser if it hits a challenge.

When a plain fetch is enough

The skill is for producing an artifact: a Markdown file, structured records, or a multi-URL result. If you only want a conversational answer about a single URL ("read this and tell me about it"), a normal page fetch is the lighter tool. Anything that yields data you keep working with belongs in WebReaper.