What is my user agent?

RFC 9110 defines User-Agent as a list of product tokens (name/version) with optional comments in parentheses. Browsers pack a lot of history into it. Here is a Chrome on Windows string, token by token:

Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0.0.0 Safari/537.36

Order matters when you sniff. Edge adds Edg/141.0.0.0 after the Chrome token, Opera adds OPR/ and Samsung Internet adds SamsungBrowser/, so a check for "Chrome" alone also matches all of them. That is why this tool uses a parser (Bowser) rather than one regex.

Browsers now freeze parts of the string to make fingerprinting harder. A parser that trusts these tokens gets the wrong answer:

When your browser supports client hints, the parsed result at the top uses them, and the notes under it say which tokens were frozen.

Chromium browsers such as Chrome, Edge and Opera expose navigator.userAgentData on HTTPS pages. The low-entropy values are there right away; the rest come from a promise:

const ua = navigator.userAgentData;
if (ua) {
  console.log(ua.brands, ua.mobile, ua.platform);
  const hints = await ua.getHighEntropyValues([
    'platformVersion', 'fullVersionList', 'architecture', 'bitness', 'model',
  ]);
  const win11 = ua.platform === 'Windows' && parseInt(hints.platformVersion) >= 13;
} else {
  // Firefox and Safari: fall back to navigator.userAgent
}

On the server, every request carries Sec-CH-UA, Sec-CH-UA-Mobile and Sec-CH-UA-Platform. To get the others, reply with an Accept-CH header and the browser adds them to its next requests:

Accept-CH: Sec-CH-UA-Platform-Version, Sec-CH-UA-Full-Version-List, Sec-CH-UA-Model
Vary: Sec-CH-UA-Platform-Version, Sec-CH-UA-Full-Version-List, Sec-CH-UA-Model

Brands like "Not)A;Brand" in the list are added on purpose ("GREASE") so servers don't hard-code the list. Ignore them.

To send a custom user agent from a script or the command line:

curl -A "MyBot/1.0 (+https://example.com/bot)" https://example.com
fetch(url, { headers: { 'User-Agent': 'MyBot/1.0' } })   // Node.js
requests.get(url, headers={'User-Agent': 'MyBot/1.0'})   # Python requests

Building a crawler? Name it and add a URL, like the bots in the sample list, so site owners can find you and allow or block you in robots.txt. The robots.txt generator writes rules per user agent, and the curl converter turns a curl command with -A into fetch, Python or Go.

MDN advises against using the user agent to decide what a page can do. Browsers lie on purpose (iPad as Mac, frozen versions), and new browsers appear. Test the feature instead:

if ('share' in navigator) { /* Web Share API */ }
if (CSS.supports('container-type', 'inline-size')) { /* container queries */ }
const touch = matchMedia('(pointer: coarse)').matches;

The user agent is still the right tool for analytics, bug reports, telling bots apart and serving the right download. For bug reports, the download button above saves the UA, the parsed result and the client hints as JSON. Testing keyboard input instead? The JavaScript keycode finder shows what each key press sends. More testing tools are in the testing tools directory.