Role: 4-tier scraping strategy.
4-tier escalation
Tier 1: WebFetch (default)
Fast, free, works for 70% of sites. Static HTML, no JS rendering.
WebFetch URL prompt="Extract main article text"
Tier 2: searxng (search-driven)
When you know the approximate target but not the exact URL. Searches first, fetches the best hit.
mcp__searxng__searxng_web_search query
mcp__searxng__web_url_read url
Tier 3: BrightData scrape_as_markdown
For Cloudflare, anti-bot, JS rendering, sites that block simple bots.
mcp__brightdata__scrape_as_markdown URL
mcp__brightdata__discover URL # for link discovery
Tier 4: Playwright (interactive)
Login-required, multi-step (click first, then extract), heavy SPAs.
mcp__playwright__browser_navigate URL
mcp__playwright__browser_snapshot
mcp__playwright__browser_evaluate "..." # for dynamic content
Decision rules
| Site type | Tier |
|---|---|
| Regular blog / Wikipedia / news article without paywall | 1 |
| Unknown URL but known topic | 2 |
| Cloudflare-protected / Reddit / X (if public) / site returning 403 to bots | 3 |
| Requires login / needs clicks / heavy SPA | 4 |
Domain-specific defaults
| Domain | Tier strategy |
|---|---|
| Government sites (parliament, ministries) | Tier 1 |
| State media with anti-bot defenses | Tier 3 |
| Telegram channels | Tier 4 (Playwright or Telegram API) |
| LinkedIn profile/post (public) | LinkedIn MCP > Tier 4 |
| Bluesky | Bluesky MCP, not scraping |
| Academic papers (arXiv, PubMed) | arxiv MCP > Tier 1 |
| Paywalled press | Tier 1 (open lede), Tier 3 if bypass via gift link |
Output format
Markdown always. Preserve source language. Add a metadata block at the top:
---
url: [original]
fetched: 2026-05-02T10:30:00Z
tier: 3
title: [extracted title]
---
[content]
Archiving
For relevant content, offer to save to the user's vault via the archive-to-vault skill (typically routes to a sources MOC or an unsorted inbox).
Version history
- v3.0 (2026-05-02): initial public release.