Making your Own CDN

I built my own CDN using Cloudflare R2 to serve media assets like .mp4, .glb, and .png—fast, clean, and from my own domain. No more using Google Drive hacks or hitting Vercel Blob's storage limits. R2 is S3-compatible, has no egress fees, and works beautifully with CLI workflows. Here's exactly how I set it up .

Why I Didn't Use Google Drive or Vercel Blob

Google Drive

It’s not made for developers.

You can upload files, sure — but try embedding a .glb or .mp4 in a site? You’ll get a redirect, a viewer wrapper, or a broken link.

• No real direct URLs

• No content-type headers

• No control

It’s a file sharing tool, not a file serving tool. Same goes for Onedrive.

Vercel Blob

• 5GB free. 100GB on Pro.

• Really good for a dedicated project or other small use cases.

But if you’re building something real — a product, a platform, a portfolio with actual media — you’ll outgrow that in a blink.

Also:

• Not S3-compatible

• No public folder structure

• Locked into their API

You can’t just curl or aws s3 cp your way in. It’s a black box with a nice UI.

Vercel Blob is designed to simplify file storage and retrieval for developers, but it doesn’t use the Amazon S3 API or protocols. Instead, it’s built on Cloudflare R2 (which is S3-compatible), but Vercel abstracts away the complexity of managing buckets, permissions, and SDK configurations. This makes it easier for frontend developers—but limits interoperability with tools or systems expecting direct S3 compatibility.

Yeah, pricing is part of it — but S3 compatibility is about freedom.

Vercel Blob is great if you want to throw a file into a black box and get a URL back. It’s clean, simple, and dev-friendly. But it’s also limited. You can’t plug it into existing infrastructure. You can’t use standard tools. You can’t automate workflows outside their SDK. It’s like building a house with no doors — looks nice, but you’re stuck inside.

I still do use Blob whenever I want to ship simple fast and efficient features/ideas.

S3 compatibility means you’re playing in the open world. You can:

• use the AWS CLI, Terraform, or any DevOps tool

• migrate to or from AWS, R2, Backblaze, MinIO, whatever

• build pipelines, backups, batch uploads, versioning, signed URLs — all with standard tools

Cloudflare R2 gives you that power — and yeah, it’s cheap. No egress fees is a game-changer. But the real win is that it doesn’t lock you in.

I know this is just a basic CDN system to allow me to transfer files — but it’s a good way. I don’t want to upgrade to Vercel Pro just for 100GB and features I’ll rarely use. I want a precise solution that fits with me.

You use Cloudflare because you’re not just building a site — you’re building infrastructure. And infrastructure should be composable, portable, and under your control.

What I Actually Needed

• S3-style uploads

• Public URLs I can embed anywhere

• Custom domain support

• No egress fees

• And a workflow that lives in my terminal

Best Options for Scalable Blob Storage

I explored several good options before settling on R2. Here’s why I didn’t go with the others:

1. Cloudflare R2

• S3-compatible

• No egress fees (huge)

• Works great with public assets

• Custom domain support

• $0.015/GB stored

• Integrates with Workers if you want edge logic

This is what I chose. And it’s been flawless.

2. Backblaze B2

Backblaze B2 is solid for backups and media storage. The pricing is attractive, and it's also S3-compatible. But it’s not optimized for serving files globally unless paired with something like Cloudflare. That means extra setup and potential latency issues. I wanted something faster and more direct out of the box.

• $0.005/GB storage

• Low egress fees

• Great for backups, media, static files

• Can pair with Cloudflare to kill egress costs

3. Supabase Storage

Supabase is a great choice if you’re already deep in their ecosystem. But I wasn’t using their auth or database, and it felt like overkill to pull in their whole stack just for file storage. I needed something more lightweight and focused purely on asset delivery.

• Dev-friendly API

• Supports public/private buckets, signed URLs

• Great if you’re already using Supabase for auth/db

• Free tier is solid; paid plans scale well

4. Firebase Storage

Firebase is popular for media-heavy apps and has tight integration with Firebase Auth. But egress costs can skyrocket, and the S3-incompatible setup makes it hard to work with standard tools. It’s too abstracted and too expensive for my use case.

• Integrates with Firebase Auth

• Signed URLs, access control, CDN-backed

• But egress can get expensive fast

My Setup: Cloudflare R2 + Custom Domain

I created a bucket on R2.

But when I tried uploading big files with the AWS CLI, I hit this:

An error occurred (InternalError) when calling the UploadPart operation

Turns out, newer versions of the AWS CLI don’t play nice with R2.

Fixing the AWS CLI Version Issue

1. Check your current version:

   Copy

   1aws --version

   shell

2. If it's anything above 2.22.35, you’ll probably hit issues.

3. I downloaded the mac ARM64 .pkg installer for AWS CLI v2.22.35 from AWS’s official site, installed it, and verified:

   Copy

   1aws --version
   2# should say: aws-cli/2.22.35 ...

   shell

Now uploads work perfectly.

Setting Up Custom Domain with Cloudflare

My domain was on Squarespace.

Cloudflare R2 only allows custom domains if Cloudflare controls your DNS.

So I:

• Added montek.dev to Cloudflare

• Changed the nameservers in Squarespace to point to Cloudflare

• Added a CNAME for assets.montek.dev

• Linked it to my R2 bucket

And just like that, I had:

https://assets.montek.dev/images/blogs/cloudflare_cdn.png

Clean. Fast. Mine.

CLI Commands I Use

1# upload a file
2aws s3 cp ./1.mp4 s3://<your-bucket-name>/videos/1.mp4 \
3
4  --endpoint-url https://<your-id>.r2.cloudflarestorage.com \
5
6  --region auto \
7
8  --profile r2

1# rename a file
2aws s3 cp s3://<your-bucket-name>/images/old.png s3://<your-bucket-name>/images/new.png \
3
4  --endpoint-url https://<your-id>.r2.cloudflarestorage.com \
5
6  --region auto \
7
8  --profile r2

1# list all files and get public urls
2
3aws s3 ls s3://<your-bucket-name>/ --recursive \
4
5  --endpoint-url https://<your-id>.r2.cloudflarestorage.com \
6
7  --region auto \
8
9  --profile r2 \
10
11| awk '{print $4}' \
12
13| while read file; do
14
15    echo "https://assets.montek.dev/$file"
16
17  done

but i didn’t want to remember all that

Shell Shortcuts in .zshrc

1r2url() {
2  echo "https://assets.montek.dev/$1"
3}
4
5r2upload() {
6  aws s3 cp $1 s3://<your-bucket-name>/$2 \
7    --endpoint-url https://<your-id>.r2.cloudflarestorage.com \
8    --region auto --profile r2
9}
10
11r2rename() {
12  aws s3 cp s3://<your-bucket-name>/"$1" s3://montek-assets/"$2" \
13    --endpoint-url https://<your-id>.r2.cloudflarestorage.com \
14    --region auto --profile r2
15  aws s3 rm s3://montek-assets/"$1" \
16    --endpoint-url https://<your-id>.r2.cloudflarestorage.com \
17    --region auto --profile r2
18}
19
20r2list() {
21  aws s3 ls s3://<your-bucket-name>/ --recursive \
22    --endpoint-url https://<your-id>.r2.cloudflarestorage.com \
23    --region auto --profile r2 \
24  | awk '{print $4}' | while read file; do
25      echo "https://assets.montek.dev/$file"
26    done
27}
28

Now I can just do:

1r2url videos/1.mp4
2r2rename images/old.png images/new.png
3r2list
4r2upload image.png images/image.png

All muscle memory.

The Result

I’ve got a zero-egress, custom-domain, CDN-backed asset pipeline.

I can upload anything, serve it globally, and never worry about bandwidth costs or ugly links.

This is how infrastructure should feel: Fast, clean, and under your control.