How to choose an anti-detect browser: 10 checks you can run yourself

To choose an anti-detect browser, test it rather than read its feature list. Ten checks cover what matters: where the fingerprint is changed, whether its values agree with each other, whether headless profiles look the same as headed ones, how proxies are checked, whether WebRTC or DNS leak, who can read your synced data, how teams share access, what automation it offers, how it keeps AI agents in bounds, and what it costs to find out. Most of these you can run in an afternoon on a free plan or a trial.

Below is each check, how to run it, and what to look for. At the end is a table with where we publish our own answers, so you can hold Browz to the same standard.

If the term is new to you, start with what an anti-detect browser is: a browser that runs separate profiles, each with its own fingerprint, proxy and storage.

1. Where is the fingerprint changed: the engine or a script?

There are two ways to make a browser report different values. One injects JavaScript into every page to overwrite properties such as the screen size or the graphics card's name. The other changes the values inside the browser engine, so the page reads them as if they were the computer's own.

Injected scripts leave traces a detector can find: a value that differs between the page and a web worker, a function whose source text no longer looks native, a property that answers differently inside an iframe. How to test: open a profile on a detector that compares page and worker values, such as CreepJS, and look for reported "lies". Then ask the vendor directly where its changes are made. Engine-level vs injected scripts explains the difference in depth.

2. Do the values agree with each other?

A fingerprint is judged less on how rare it is than on whether it hangs together. A Mac user agent with Windows fonts, a timezone that does not match the IP address, or a screen size that CSS and JavaScript report differently all give a profile away. Random noise added to canvas or audio is another tell, because detectors read a value twice and compare.

How to test: run a consistency test in a fresh profile on default settings, and read every row, not only the score. Then open three profiles on a device-identification test and check that they are seen as three separate visitors, each the same after a restart. Our own browser fingerprint test runs in your browser and sends nothing anywhere.

3. Does a headless profile look the same as a headed one?

If you plan to automate, profiles will often run without a window. Many browsers report different values headless: a different user agent, missing plugins, an odd screen. How to test: run the same consistency test with the profile headed, then headless, and compare the two results line by line.

4. Are proxies checked, and does the profile follow the exit?

One proxy per profile is the minimum. What separates tools is what happens next. How to test:

1. Add a proxy and run the browser's own check. Does it open a real tunnel and measure the exit's address and country, or only test that the port answers?
2. Start the profile and compare its timezone and languages with the exit's location.
3. Change the proxy's exit to another country and start the profile again. Does the browser notice, warn or refuse?

5. Do WebRTC or DNS leak?

WebRTC, the technology behind video calls, can reveal your real network address past a proxy. DNS lookups can go out through your own network instead of the proxy. How to test: open a WebRTC leak test page and a DNS leak test page in a profile with a proxy. Only the proxy's exit should appear.

6. Who can read your synced profiles?

If profiles sync between computers, they pass through the vendor's server: cookies, sessions, proxy passwords. How to check: read the security page for one sentence that matters, whether data is encrypted on your device before it leaves, with a key the server does not hold. "Encrypted in transit" or "encrypted at rest" alone means the vendor can still read it. Also check how exported profiles are protected.

7. Can a team share access without sharing passwords?

For agencies and teams, look for roles, sharing at several levels (view, launch, edit, manage), a rule that a shared profile opens on one computer at a time, a log of who launched or changed what, and a clean way to transfer a profile when someone leaves.

8. What automation does it offer, and does automation change the fingerprint?

Check for a local API, a command line, CDP access for Playwright, Puppeteer and Selenium, SDKs, and an MCP server for AI agents. Then the important part: attach Playwright to a profile and run the consistency test from step 2 again. Driving a profile should not change what it presents. Playwright in a real browser profile shows the attach pattern.

9. How does it keep AI agents in bounds?

If agents will use the browser, ask what stops an agent from going further than its task. Look for:

• a key per agent, limited to chosen profiles and sites, that you can revoke;
• sign-ins an agent can use without reading the password;
• a way for the agent to hand a login, 2FA code or CAPTCHA to a person;
• a log of every call, kept on your own computer;
• published, repeatable measurements of agents finishing real tasks.

10. What does it cost to find out?

A fair test needs a few profiles and a little time. Check whether there is a free plan or trial with enough profiles to run the checks above, what the cheapest paid plan costs, and whether the first payment is refundable if the browser does not suit you.

A test plan for one afternoon

1. Install the browser and create three profiles on default settings, one with a proxy.
2. Run checks 1–3 on all three: a detector for lies, a consistency test headed and headless, a device-identification test.
3. Run checks 4 and 5 on the profile with the proxy.
4. Read the security and pricing pages for checks 6 and 10.
5. If you automate, attach Playwright and repeat the consistency test (check 8).
6. If you use agents, connect one through MCP and try to make it open a site outside its key (check 9).
7. Write down each result with the date, because detectors and browsers both change.

Our own answers

We publish ours so you can check them, and the free plan is enough to repeat most of them.

For a side-by-side look at the browsers that agent builders consider, see our comparison of browsers for AI agents. Whatever you choose, use it only where each site's terms allow; our acceptable use policy sets out how we apply that.

Frequently asked questions