BGP prefixPrefix
The block of addresses this IP is announced in to the rest of the internet. RPKI valid means the block's owner
signed off on the network announcing it.
RPKI valid
172.66.40.0/21
Seen by 3,425 peers
Latency
Round trip for an HTTPS request from your browser to the site and back, including the server's reply time.
Live
— ms
Measuring from your browser…
At a glance
Route
From Columbus on Anthropic, PBC
YouColumbus
Finding a probe near you…
Networks crossed
Each network is an autonomous system (AS). Your traffic is handed from one to the next where they connect, following
the routes BGP advertises.
AS16509Anthropic, PBC
AS13335Cloudflare
#HostNetworkLocationRound trip
AS
⋮
Waiting for a reply——*
Tracing from a probe near you with Globalping. Hop locations that its round-trip times rule out are left off the map.
Email
Protected from spoofing
Mail that fails its checks is sent to spam under clickhouse.com's DMARC policy, so forged mail doesn't reach inboxes.
Receives mail
Google Workspace
Mail servers
aspmx.l.google.comalt1.aspmx.l.google.comalt2.aspmx.l.google.com+2 more
SPFLists the servers allowed to send mail for the domain. Receivers check it against the sending server.
Soft-fails everything else6 of 10 DNS lookups · includes _spf.google.com, mktomail.com…
DMARCTells receivers what to do with mail that fails SPF and DKIM: nothing (none), spam folder (quarantine) or reject.
Quarantinesends reports
MTA-STSRequires other mail servers to use encrypted connections (TLS) when delivering to this domain.
Not published
TLS reportsAsks other mail servers to report failed encrypted deliveries (TLS-RPT).
Not requested
BIMIA brand logo that some mail apps show next to authenticated mail from the domain.
ClickHouse is a fast open-source column-oriented database management system that allows generating analytical data reports in real-time using SQL queries
Language
American English
Built with
CloudFrom the provider's published address ranges.
Built with
Each entry says what gave it away: a response header, a cookie's name, an asset's path or file name, or markup the tool leaves in the page.
Versions are checked against endoflife.date's support schedules and, for JavaScript packages, OSV's vulnerability database, which carries
GitHub's security advisories.
Hosting
Vercel
x-vercel-id header
CDN & edge
Cloudflare
cf-ray header
Framework
Next.jsApp Router
x-nextjs-prerender header
UI
React
Next.js renders with React
Styling
Tailwind CSS
187 Tailwind CSS class names
Fonts
Basiersquare
self-hosted font
Soehne Breit Buch S P
self-hosted font
Soehne Breit Extrafett S P
self-hosted font
Soehne Breit Fett S P
self-hosted font
Soehne Buch S P
self-hosted font
Soehne Dreiviertelfett S P
self-hosted font
Soehne Extrafett S P
self-hosted font
Soehne Fett S P
self-hosted font
Components & icons
Lucide
8 Lucide class names
Bot protection
Cloudflare Bot Management
__cf_bm cookie
How it's made
RenderingPrerendered pages are built ahead of time and served as files. Pages rendered on each request are built by server code for every visitor. Pages drawn in the browser arrive nearly empty and depend on JavaScript, which most AI crawlers don't run. Islands are interactive parts in otherwise static HTML.
Prerenderedx-nextjs-prerender headerApp Router
Edge cacheWhether a CDN or proxy answered heade.rs's request from its cache, and how long ago that copy was made (the Age header).
Not cachedCloudflare · made 33 hr ago
Cache-Control
public, max-age=0, must-revalidate
HTML
343.3 KB
Data for scriptsData the server writes into the HTML so JavaScript can pick up where the server left off (hydration). Browsers download and parse all of it before the page responds, so large payloads slow pages down, even when the page looks ready.
React Server Components 189.1 KB · 55% of the HTML
Scripts
26 files · 7 inline2 from other sites
Render-blockingPlain scripts in the page's head, without async, defer or type=module. The browser shows nothing until each one has downloaded and run.
1 script
Code from other sites
Scripts a page loads from another domain run with the same power as its own code: they can read what visitors type and change the page.
heade.rs checks what the HTML loads, not what tag managers add once they run.
Integrity checksSubresource Integrity: the script tag carries a hash of the file, so the browser refuses it if the other site changes it. Analytics and tag manager scripts change by design and can't carry one; pinned library files can.
robots.txt lets every AI crawler in. Agents that ask for Markdown get 20.7 KB instead of 343.3 KB of HTML, and it offers an MCP server, an API catalog and 10 agent skills.
Crawlers
Allows all
TrainingSearchOn request
They visit a page when someone asks an assistant about it. ChatGPT's, Perplexity's, Meta's and Amazon's treat robots.txt as advisory;
Anthropic's and Mistral's follow it.
Can cite itThe AI products that can use the site in their answers: each answers from one crawler (OAI-SearchBot for ChatGPT search, Googlebot for AI Overviews, bingbot for Copilot), and robots directives on the front page, like noindex, nosnippet and noarchive, keep some out.
Content signals
Lines in robots.txt (Cloudflare's Content Signals Policy) saying whether pages may be used for search, as input to AI answers (ai-input) and for AI
training. No major AI company has said it follows them.
ai-train=yessearch=yesai-input=yes
It's on Cloudflare, which can also block AI crawlers at its edge whatever robots.txt says.
Reading
Without JavaScriptText on the front page before any scripts run: what GPTBot, ClaudeBot, PerplexityBot and most agents read. Google, Bing and Apple's crawlers run scripts.
1,286 words
MarkdownAgents like Claude Code, Cursor and OpenCode ask for Markdown in their Accept header. It's the page without its markup, so it costs them far fewer tokens.
On request20.7 KB, 17× smaller · converted by Cloudflare
llms.txtA Markdown file listing a site's key pages for AI tools. Coding agents read it; search crawlers mostly don't.
Structured dataschema.org JSON-LD in the page: facts about the site in a form machines read directly.
WebSite
Tools
Ways an AI agent can do things on the site, not just read it: an MCP server that apps like Claude, ChatGPT and Cursor connect to, an API
catalog, packaged agent skills, a shopping profile (UCP), or tools the page offers AI in the browser (WebMCP). heade.rs looks in the official
MCP Registry and the files the site publishes under /.well-known/.
MCP serverModel Context Protocol: a server that AI apps like Claude, ChatGPT and Cursor connect to so they can use a service's tools. Servers in the official MCP Registry are published under the domain's own name, which the registry checked. A server card, which the site's AI Catalog points to, says which MCP revisions the server speaks; 2026-07-28 is the first stateless one.
ClickHouse Cloud MCP Servermcp.clickhouse.cloud
APIsAn API catalog (RFC 9727, at /.well-known/api-catalog) lists a site's APIs and links to their machine-readable descriptions, usually OpenAPI.
Agent skillsInstructions and scripts packaged for AI agents (SKILL.md), listed in an index at /.well-known/agent-skills/index.json or in the site's AI Catalog. Tools like Vercel's skills CLI install them.