Web interface

The Compose service serves this frontend beside its API. The Cloudflare read Worker serves its own static asset bundle.

web/ holds the frontend with no build step: index.html, app.css, app.js, robots.txt, site.webmanifest, and the icon set. ctlogs.web.mount_frontend registers each one as an explicit named route at startup, so the page and the API share an origin. They are named routes rather than a StaticFiles mount because create_app mounts the MCP app at /, and a second directory mount on that prefix would swallow /mcp. Only those names are servable, so a stray file dropped in web/ is not reachable.

Subfinder serves ASSETS with Cache-Control: no-cache, because the markup and the assets ship together and a browser must not pair new markup with a cached script. The icons in MEDIA carry no such pairing, so they are cached for a week instead. Brand sources live in brand/, which is not served.

robots.txt disallows /v1/, /mcp, and the ?apex= form of the page. Those routes spend the shared search allowance, and a crawler following every result link would spend a visitor's day of reads on nobody's behalf. The page carries rel="canonical" pointing at / for the same reason: /?apex=example.com is the same document with a query on it, not a second page.

The page is optional. When web/index.html is missing, mount_frontend logs and returns, and the API serves alone. Set CTLOGS_WEB_DIR to serve the frontend from another directory. The Dockerfile copies web/ after the dependency install so static edits do not invalidate that layer.

Opening the page spends nothing from the search allowance. A visitor spends a read only when they search, and a repeat lookup of an apex already read in that browser session comes from memory instead of the API. The counter in the page header polls /v1/stats every 15 seconds and pauses while the tab is hidden. That polling is only safe while /v1/stats stays outside the allowance. At a 15 second interval a metered stats route would spend all 1,000 daily reads in about four hours for a visitor who never ran a search. test_opening_the_page_spends_no_search_allowance and test_the_live_counter_endpoint_spends_no_search_allowance in tests/test_web.py pin both, so adding a consume() call to /v1/stats fails the suite rather than quietly draining callers.