setup-tools

setup-tools

One-time (re-runnable) setup for the Local-Web-Capture plugin. Idempotent — safe to run repeatedly.

Why this plugin exists

All captures run from the user's localhost so the HTTP request originates from the user's Israeli IP. This is deliberate — it is the reason we do not use hosted readers like Jina or Firecrawl SaaS for these jobs.

Steps

1. Verify data directory. Check ~/local-web-capture/ exists with subdirs articles/, prices/, translations/, .cache/. Create any missing. Confirm sites.yaml exists at the root — if missing, copy from the plugin's reference seed.

2. Install primary extractor: Scrapling. Covers Tier 1 (static), Tier 2 (stealth via Camoufox), and Tier 2b (Playwright) in one Python lib.

   pip install --user scrapling
   scrapling install    # fetches Camoufox + Playwright browser binaries
   

   Verify:

   python -c "from scrapling.fetchers import Fetcher; print(Fetcher.get('https://example.com').status)"
   

3. Install Tier-3 extractor: bb-browser. For the authenticated/real-Chrome path.

   • Repo: https://github.com/epiral/bb-browser
   • Install per upstream docs, then register as an MCP server if the plugin's scrape-authenticated skill will call it via MCP.
   • Confirm the user is signed in to target paywalled sites (Haaretz, Calcalist+) in their real Chrome profile.

4. Install backups (optional, only if a site defeats Scrapling):

   • webclaw — Rust CLI Tier-1 fallback. cargo install webclaw or prebuilt binary to ~/.local/bin/webclaw.
   • camofox-browser / stealth-browser-mcp — Tier-2 fallbacks.
   • Playwright MCP — Tier-2b fallback (may already be installed).

5. Optional translator fallback: LibreTranslate. Only if the user plans bulk translate jobs. Not required for occasional use — Claude handles inline.

   pip install libretranslate
   

6. Playwright MCP transitional note. The existing user-level Playwright MCP is a usable Rung-2 fallback during bring-up. Once webclaw + bb-browser are wired and validated on real captures, remove Playwright from user-level MCP config (~/.claude/settings.json) so Claude does not bias toward it. Leave it as a project-level MCP if still needed occasionally.

7. Report status. Print a one-screen summary: data dir path, Scrapling version (+ whether scrapling install has been run), bb-browser status, backups present (webclaw / camofox / stealth-mcp / Playwright MCP), sites.yaml entry count.

Notes

• Do not run captures from this skill — it only sets up the environment.
• If a dependency install would take >60s, tell the user and proceed; do not wait silently.
• Never write API keys or credentials into the data directory.