​

Agent Decision Guide

webrain exposes 15 intent-based MCP tools. This page is a decision table: given a task, which tool (and which what / action / mode / op selector) should your agent call? The same rules are served to the LLM at runtime by webrain_guide.

​

The 15 tools at a glance

​

Decision tree

I need to reach a page → webrain_navigate. Read the returned challenge field — if non-null, the page is gated (see Challenges). I need to see what’s on the page → webrain_observe:

• what=state / a11y / semantic — DOM structure for reasoning
• what=html / images — raw content
• what=screenshot — visual, or what=pixel to inspect coordinates

I need to click / type / scroll → webrain_interact (action=click, type, press, scroll, select, hover, check, upload, wait, dismiss_overlays, …). I need data from the page → webrain_extract:

• mode=autoschema — no selectors; probes the DOM and proposes a schema
• mode=schema — extract with your JSON schema
• mode=regex / table / jsonld / bm25 — targeted modes

Static HTML, no JS? → webrain_scrape (10–100× faster than a browser). One URL → many pages → webrain_crawl (mode=spider, sitemap, scan). A fixed list of URLs → webrain_batch (op=fetch | extract | interact | eval | screenshot, optional cdp_urls fan-out). Logged-in / gated content → webrain_session (op=open → login → op=cookies / setcookies, profile save/restore). See Auth & Login. A PDF, a video, a file → webrain_pdf / webrain_watch / webrain_download. Screenshots or visual QA at scale → webrain_vision (op=index / retrieve).

​

Task → tool quick map

​

Related

• Tools Reference — exact schemas and legacy aliases
• Browser Engines — which browser for which job
• Structured Extraction
• Scraping at Scale
• Auth & Login