FREEFirst project with us? Your landing page design is on the house.See the offer

Chapter 2 of 6

Reverse proxy setup in plain words

Three hostnames, one route table and a few settings that decide if it works. How to plan the setup, what Webflow asks for, and how to test it before real traffic arrives.

7 min readUpdated 2026-11-07
All chaptersChapter 2 of 6
  1. 01When a full cutover is wrong
  2. 02Reverse proxy setup in plain words
  3. 03SEO parity for each page
  4. 04Forms, tracking and sign-in across two systems
  5. 05Moving platforms in phases
  6. 06Retiring the old site

The short answer

Point your public domain at the proxy. Give each system its own origin hostname, such as wf.example.com and old.example.com. Then write a route table that says which paths go where, with the old site as the default until most pages have moved. Everything else in this chapter is about making that hold up.

The routing idea itself is in the short post. Here we go through the setup.

Three hostnames, not two

Once the public domain points at the proxy, each system needs another address the proxy can reach.

HostnamePoints toWho should see it
www.example.comThe proxyEveryone
wf.example.comWebflowOnly the proxy and your team
old.example.comThe old hostOnly the proxy and your team

The old site is easy to forget here. It needs to answer on its new hostname, while still writing links with the public domain. Many CMSs have a site address setting that controls this, so check what yours does before you change it.

Do not leave these names to be sorted out on launch day. Set up and test the origins first, with the public domain still pointing where it does today.

What Webflow asks for

From Webflow's self-managed proxy article and its reverse proxy checklist:

  • A paid Site plan and a connected domain.
  • An origin subdomain, such as wf.example.com, set as the default domain.
  • A global canonical set to your public URL, also used as the base for your sitemap.
  • No proxying of *.webflow.io. Webflow says that is not supported and breaks its terms. It also stops that address working as your staging site.
  • SSL on the origin. The checklist says wildcard certificates from Cloudflare are not compatible, so you need certificates for each explicit site name.
  • The real visitor IP sent in X-Forwarded-For. Otherwise Webflow sees all traffic coming from one IP, which can look like an attack and get blocked.

These details change, so check the current docs before you build. If you are on Enterprise, Webflow directs you to Sales.

Write the route table first

Write this before you write any proxy code. It is your plan, your test list and your record. Here is an example for a site in its first batch.

PathGoes toBatchNote
/ (exact)Webflow1Exact match, or it catches everything
/pricingWebflow1
/aboutWebflow1
/robots.txtOne chosen systemSetupCrawlers read just one file
/sitemap.xmlOne chosen systemSetupCan be an index of both
Everything elseOld siteDefaultIncludes the old site's files and scripts

Keeping the old site as the default means a mistake shows an old page, and that is rarely a disaster. It also means the old site's images, scripts and plugin files keep working without a rule each. When you later switch the default to the new system, you will need rules for every one of those paths, so log them now.

Decision rule Switch the default only when the new system holds most of the pages and you have a list of every old path that must still go to the old site.

Rule patterns that catch people out

Pattern rules differ by tool, so read the docs for yours. Cloudflare's Workers routes are a clear example. A route maps a URL pattern to a Worker, and the Worker can act as a proxy in front of your server. Its docs say:

  • * is the only operator, and it cannot sit in the middle of a path, so example.com/*.jpg is invalid.
  • When several patterns match, the most specific one wins.
  • Paths are case sensitive.
  • Patterns cannot include query parameters, so URLs with query strings are matched only through a trailing *.
  • A route with no Worker attached switches off a less specific route for matching requests.
  • The domain needs a proxied DNS record.

Two habits follow from this, and they apply to most tools. A rule ending in /blog/* leaves /blog uncovered, so test both. A rule ending in /blog* also catches /blogger, so check nothing else starts with those letters.

Things the proxy must pass along

  • Query strings. Campaign tags, site search and CMS pagination all depend on them. The checklist calls out all three, so test each.
  • Redirects. If an origin answers with a redirect to wf.example.com, the visitor lands on the origin. Rewrite those addresses to the public domain, and fix the cause.
  • Trailing slashes. Webflow removes them. The checklist says your proxy should match, to avoid redirects to the origin.
  • Status codes. A 404 from an origin must stay a 404. If the proxy turns errors into a friendly 200 page, you create soft 404s.
  • The right cookies. Chapter 4 covers this, since sending every cookie to both systems has a cost.

Caches and script deferral

You now have more than one cache. Webflow clears its own when you publish, but it cannot clear yours. Webflow's article gives two choices: turn off your CDN's caching, or plan to purge both. The checklist adds that the site_publish webhook can trigger your purge automatically, so no one has to remember.

If query strings change what a page shows, make sure the cache treats them as different pages. Never cache a response that sets a cookie or depends on a signed-in visitor.

Also turn off script deferral features on your CDN. The checklist names Cloudflare Rocket Loader, which can break interactions, forms, sliders and custom code.

Test before real traffic

You can test the whole setup on a spare hostname first, such as proxy.example.com, with a noindex header so it stays out of search. Then run this list against it, and again on the public domain.

  1. Request every path in the route table and confirm which system answered. A temporary response header, such as x-served-by: webflow, makes this quick with curl -I.
  2. Request a page that does not exist under each system, and confirm a real 404.
  3. Search the page source for your origin hostnames.
  4. Send a request with a query string and confirm it arrives intact.
  5. Publish a small change and time how long it takes to show.
  6. Turn the proxy's route for one page on and off, to rehearse a rollback.

Decide who gets the alert when it fails, too. Google's guidance on server errors says crawling slows down on 5xx responses, and URLs that keep returning them are eventually dropped from the index. An outage nobody notices is the real risk, so watch the public domain from outside, on a few URLs from each system.

Takeaways

What to remember from chapter 2.

  • Each system gets its own origin hostname, and the public domain points at the proxy.
  • Webflow asks for a paid Site plan, a connected domain, an origin subdomain and a global canonical, and rules out proxying webflow.io.
  • Write the route table first, and keep the old site as the default until most pages have moved.
  • Pass query strings, status codes and the real visitor IP, and rewrite redirects to the public domain.
  • Plan your cache purge, turn off script deferral, and watch the public domain from outside.

Your turn

Let's kook
what you krave.

Hi Kookies, I need a built on . I'd love to launch . Reach me at
or book a callSent. We reply within one working day.