Not an HTTP client. A real browser.
A real browser, not just `fetch`
`browser_open` / `browser_navigate` / `browser_search` drive a real Chromium instance — JavaScript executes, cookies persist, sessions survive. That is the difference between "scrape a dynamic site" and "LLM writes a scraper."
Closed platforms like Xiaohongshu
Optimized for sites without open APIs: session keep-alive, anti-detection tactics, content extraction baked in.
Properly sandboxed
The agent’s browser is isolated from your daily browser — no profile pollution, no cookie cross-contamination. Clear or disable it independently.
Wired into the rest of the agent
Scraped content can flow into `web_article_save` (lands in the vault as Markdown) or into `apply_patch` to append to a draft you are writing.