site-analyser

Site-quality signals for a deployed URL or a local static-site directory — the lens-family member that reads a rendered/deployed website rather than its source files.

code-analyser reads .html/.css/.js source files in isolation; this one crawls a site (live or local), parses each page, and returns accessibility / structure / SEO / link-health / framework / validity / perf signals. Pure-Python core (always pip-installable); optional Lighthouse and the W3C Nu HTML Checker for deeper signals when those tools are on PATH (graceful degradation otherwise).

Install

pip install site-analyser

Optional deeper signals (no Python install needed — site-analyser shells out if it finds them):

# Lighthouse (perf / a11y / SEO / best-practices scores) — needs Node + Chrome
npx -y lighthouse --version

# W3C Nu HTML Checker (deep HTML/CSS validation) — needs Node + Java
npx -y vnu-jar --help

Use

Python:

from site_analyser import SiteAnalyser

result = SiteAnalyser().analyse(url="https://example.com")          # live URL
result = SiteAnalyser().analyse(path="./build")                     # local static-site dir
print(result.overall.page_count)                                    # 12
print(result.overall.accessibility.alt_text_coverage)               # 0.94
print(result.overall.broken_links)                                  # []
print(result.overall.frameworks_detected)                           # {'bootstrap': ['/index.html']}

CLI:

site-analyser https://example.com                # human summary
site-analyser ./build --json                     # raw JSON of a local dir
site-analyser https://example.com --max-pages 20
site-analyser ./build --no-external              # skip Lighthouse/vnu even if available
site-analyser serve                              # HTTP API on port 8012
site-analyser manifest                           # capability manifest

HTTP (site-analyser serve on port 8012):

curl -X POST http://localhost:8012/analyse \
  -H 'content-type: application/json' \
  -d '{"url": "https://example.com", "max_pages": 10}'

curl http://localhost:8012/health
curl http://localhost:8012/manifest

Like git-analyser, the API takes a JSON body (not a multipart upload) — exactly one of url or path must be set.

Signals

Per page and rolled up across the site:

• Crawl — internal-link discovery, page count, broken links (internal HEAD checks).
• Structure — semantic HTML (<header>/<nav>/<main>/<footer>), heading hierarchy (missing <h1>, skipped levels, depth).
• Accessibility (WCAG heuristics) — alt-text coverage, form-label coverage, lang on <html>, ARIA usage, skip-link, image count, doc-language coverage.
• SEO — <title>, meta description, viewport, canonical, OpenGraph coverage.
• Tech — framework detection (Bootstrap, Tailwind, Bulma, Materialize, React, Vue, Angular, jQuery, Svelte), CDN-link detection, inline style= / <style> / <script> / on*= counts.
• Validity — HTML parse-error count (html5lib); deep W3C via vnu if present.
• Perf — page weight (bytes), HTTP status (URL mode); real Lighthouse scores if present.
• External tools (when on PATH) — Lighthouse categories (perf/a11y/SEO/best-practices), vnu HTML/CSS error/warning counts. Absence is reported (not an error).

The family

Part of the lens analyser family.

Limits

• Pure-Python a11y/perf signals are heuristics — Lighthouse remains the gold standard for the full picture. Install it (npx lighthouse --version works) for the deep numbers.
• The crawler follows same-origin links only, up to max_pages (default 10).
• Broken-link checking does internal HEAD requests; external links are not fetched by default.
• v1 doesn't handle JS-rendered content — pages are parsed from the raw HTML response.

License

MIT