Short answer: S3 is an API, not a protocol. Amazon Simple Storage Service (S3) is a storage service from AWS. You use it through the S3 API, which runs over HTTP or HTTPS. Those are the actual network protocols. “S3 protocol” is a common shorthand, but there is no formal S3 protocol standard, the way there is for FTP, NFS or SMB.
Service, API and protocol: what the terms mean
Three terms get mixed up here. This table separates them:
| Term | What it means | S3 example |
|---|---|---|
| Service | A product you use, with its own features and billing | Amazon S3 |
| API | The operations and rules for requesting the service | PutObject, GetObject, ListObjectsV2 |
| Protocol | The rules for sending messages across a network | HTTP and HTTPS |
Put together, the layers look like this. Your code calls the S3 API. The API sends HTTPS requests. The S3 service answers with object data or XML responses, depending on the operation.
How requests to S3 work
The core S3 operations include:
- PutObject uploads an object.
- GetObject downloads an object.
- HeadObject reads an object’s metadata without downloading the data.
- DeleteObject removes an object.
- ListObjectsV2 lists keys in a bucket, optionally filtered by prefix.
- Multipart operations (CreateMultipartUpload, UploadPart, CompleteMultipartUpload) upload large objects in parts.
Most people don’t call these directly. The AWS CLI and the AWS SDKs wrap them. This command uploads a file:
aws s3 cp report.pdf s3://example-bucket/reports/report.pdf
Behind that command, the tool sends an HTTPS request to the S3 API. A raw request looks roughly like this, with the authorization header shortened:
PUT /reports/report.pdf HTTP/1.1
Host: example-bucket.s3.us-east-1.amazonaws.com
Authorization: AWS4-HMAC-SHA256 Credential=... (abbreviated)
Content-Length: 204800
Requests to S3 are signed with AWS Signature Version 4 (SigV4). Signature Version 2 is deprecated. S3 once also offered a SOAP interface, but it is deprecated, and REST is the supported interface.
Why people call it a protocol
Several things make “S3 protocol” feel natural:
- The
s3://address form. Tools such as the AWS CLI uses3://bucket/keyas a naming convention. It is not a registered network protocol. Browsers and operating systems don’t understand it natively. - The API became the standard. S3 became the de facto standard for object storage, so many systems implement the same API. That makes the API feel like a protocol.
- Compatibility. Other providers offer S3-compatible storage, so applications can talk to them using the same requests.
What S3-compatible storage means
Many services implement the S3 API, including MinIO, Ceph RADOS Gateway, Wasabi, Backblaze B2, Cloudflare R2, DigitalOcean Spaces, IBM Cloud Object Storage and the XML API of Google Cloud Storage in its interoperability mode.
“S3-compatible” does not mean identical. Advanced features such as Object Lock, S3 Select, some lifecycle options and IAM-specific policies may be missing or behave differently. If your application relies on a feature, test that feature against the specific provider. For more on how object storage providers compare, see object storage in cloud computing.
Is S3 a file system?
No. S3 stores objects, and you cannot edit a file in place the way you can on a local disk. You can still present S3 to file-based applications in a few ways:
- AWS Storage Gateway (S3 File Gateway) presents S3 buckets as NFS or SMB file shares to on-premises systems.
- Mountpoint for Amazon S3 is a FUSE client that mounts a bucket on Linux. It has limits, such as no in-place edits of existing objects.
- Open-source tools such as s3fs and rclone can also mount buckets, with their own limits.
NFS and SMB are true file-sharing protocols. iSCSI is a block protocol. S3 belongs to neither group, which is why these tools exist as adapters.
Where to go from here
- If you write code, use the AWS SDK for your language. It wraps the S3 API and handles signing.
- If you run scripts or manage buckets, the AWS CLI is the usual tool.
- If you are choosing a service, read Amazon S3 explained for buckets, storage classes and defaults.
- For the wider category, see types of cloud storage services. For block-versus-object trade-offs, see why use object storage instead of block storage?.