perusal

The internet without all the 💩

pip install perusal

perusal https://en.wikipedia.org/wiki/Domino    # -> clean Markdown on stdout
perusal ./saved-page.html                        # a local HTML file works too
perusal report https://example.com               # HTML-vs-Markdown size / token savings
perusal url1 url2 url3                           # -> one combined Markdown digest
echo '<h1>hi</h1>' | perusal -                   # read HTML from stdin

Pages pass through the domonic Readability and Turndown ports: fetch → readable-article extraction → Markdown → cleanup, with optional relevance ranking (--query) and a hard character budget (--max-chars).

Pages that render client-side (an empty server response, "Enable JavaScript to view this content") need their own JavaScript actually run first:

pip install 'perusal[js]'
perusal https://example.com/app --js             # runs the page's <script>s via myjs first

The reader-browser

pip install 'perusal[browser]'
perusal browse                                   # tabbed webview reader
perusal browse https://en.wikipedia.org/wiki/Domino

A small tabbed reader-browser (history, incognito tabs, bookmarks, DevTools, adjustable type, Wikipedia math rendering, "save page as Markdown") with keyboard shortcuts: ⌘/Ctrl+T new tab, W close tab, R reload, D bookmark, B toggle bookmarks, U view page source, L focus the URL bar.

DevTools (the </> button): raw page Source, a static Resources inventory (every referenced script/stylesheet/image/iframe, each viewable in turn), and an Info tab with IP / reverse DNS / WHOIS / DNS records (hand-rolled clients, no new dependency) plus certificate-transparency-log subdomain discovery via crt.sh. Purely passive and manually triggered — no port scanning, no active brute-forcing, nothing sent to the site itself beyond the page fetch perusal already made.

MCP server (optional)

pip install 'perusal[mcp]'
perusal mcp                                       # stdio MCP server

One read-only, idempotent tool — web_read(url, query?, max_chars?, timeout?) — built on the official Python MCP SDK. Point an MCP client at perusal mcp.

See the full documentation.