memgrafter/website-scraper

Website Scraper

Archive web pages into a personal knowledge base with structured summaries and preserved source text.

Why use it

• Preserve disappearing pages — keep a local copy before content changes or vanishes
• Fast recall — read a concise summary instead of re-reading the full page
• Reliable re-use — raw extracted text is saved for future LLM context
• Searchable metadata — YAML frontmatter supports filtering/query workflows

When to use

• Saving tutorials, docs, blog posts, or repo READMEs for later
• Building a research/reference archive
• Capturing sources for ongoing project work
• Converting “read later” links into durable notes

When NOT to use

• JavaScript-heavy SPAs that require browser rendering
• Pages behind auth/paywalls/captchas
• High-volume crawling (this flow is designed for one URL at a time)

Usage

./run.sh "https://example.com/article"

Examples

# Archive a blog post
./run.sh "https://simonwillison.net/2024/Dec/19/one-shot-python-tools/"

# Archive a GitHub README page
./run.sh "https://github.com/SWE-agent/mini-swe-agent"

# Use a custom archive location
DATA_DIR=~/research ./run.sh "https://arxiv.org/abs/2501.09891"

Output

Per URL, the scraper writes:

• {year}/{date}_{slug}.md — summary + YAML frontmatter
• {year}/raw/{date}_{slug}.txt — raw extracted text (for LLM context)
• {year}/README.md — auto-updated index row

How it works

Fetch page → extract clean text (trafilatura) → generate validated summary/frontmatter → save summary + raw text.

Cost: ~0.5 cents/page. Benefit: permanent, searchable, LLM-ready references.