docs/gateway.md
An IPFS Gateway acts as a bridge between traditional web browsers and IPFS. Through the gateway, users can browse files and websites stored in IPFS as if they were stored in a traditional web server.
More about Gateways and addressing IPFS on the web.
Kubo's Gateway implementation follows IPFS Gateway Specifications and is tested with Gateway Conformance Test Suite.
By default, Kubo nodes run
a path gateway at http://127.0.0.1:8080/
and a subdomain gateway at http://localhost:8080/.
[!CAUTION] For browsing websites, web apps, and dapps in a browser, use the subdomain gateway (
localhost). Each content root gets its own web origin, isolating localStorage, cookies, and session data between sites.For file retrieval, use the path gateway (
127.0.0.1). Path gateways are suited for downloading files or fetching verifiable content, but lack origin isolation (all content shares the same origin).
Additional listening addresses and gateway behaviors can be set in the config file.
For public gateways available as a public good, see the public utilities page; for a broader community-maintained list, see the public gateway checker.
Treat public gateways as a convenience for casual use; they are provided on a best-effort basis. For anything you depend on, run your own gateway (recipes below) or retrieve content over IPFS directly instead of through someone else's gateway. The guide Replace public gateways with self-hosted IPFS covers the options.
Before you pick a URL style (subdomain, path, or DNSLink, in the URL-style recipes below), decide the bigger question: when a visitor asks for content this node does not have, should the gateway go and fetch it from the network, or just answer "not found"?
[!IMPORTANT] See Reverse Proxy Caveats if running behind nginx or another reverse proxy.
[!CAUTION] An open public gateway fetches and serves any content a visitor asks for, even content you have never seen. That means a stranger can pull illegal or abusive material through your server and your internet connection, and you are the one who has to handle the complaints and takedown requests. Unless you are running a shared public service on purpose, it is safer to serve only your own content with
Gateway.NoFetch=true. If you do run an open public gateway, first set up a way to block and take down bad content, using content blocking.
Gateway.NoFetch=true)The node only shares content it already has. It never downloads missing content
from other peers, so anything this node does not have returns 404 Not Found.
This is the safest way to put your own content (a website, a dataset, some files) on a public gateway, and the recommended setup for most people.
$ ipfs config --json Gateway.NoFetch true
ipfs add, pinned, or put in MFS. Anything else returns
404 Not Found right away, without touching the network.ipfs pin add) so garbage collection does not
remove it. Unpinned content can be deleted when garbage collection runs, after
which it stops resolving and returns 404.Provide.Enabled=false. That drops the ongoing
announcing work.NoFetch does not stop your node from helping other peers find their
content. If you want it to act only as a client and not help route for others,
set Routing.Type=autoclient.Gateway.NoFetch=false)This is the default, and how a large shared public gateway works. The gateway looks up and downloads any content a visitor asks for, fetching it from the network. Only run it this way if you mean to offer that kind of public service, and set it up carefully first.
410 Gone for it. Blocking stops your node from serving the content, but
not from helping other peers find it
on the network; to stop that too, set
Routing.Type=autoclient.Gateway.DeserializedResponses=false so the
gateway returns only verifiable data, not ready-to-view web pages. People can
no longer use it to host random websites, while apps that verify data
themselves (like @helia/verified-fetch)
keep working.Gateway.MaxRangeRequestFileSize,
Gateway.MaxConcurrentRequests, and the
Running in Production notes on timeouts,
reverse proxy, and CDN behavior.Routing.AcceleratedDHTClient=true
for faster lookups. Decide whether this gateway should also announce the
content it fetches so others can find it: if yes, turn on
Provide.DHT.SweepEnabled=true (and raise
Provide.DHT.MaxWorkers if announcements are slow);
if no, set Provide.Enabled=false.These decide how a request URL maps to content. They work the same whether or
not Gateway.NoFetch is set.
A subdomain gateway
serves each content root from its own subdomain
(http://{cid}.ipfs.subdomain-gw.example.com), so every root gets its own Origin.
$ ipfs config --json Gateway.PublicGateways '{
"subdomain-gw.example.com": {
"UseSubdomains": true,
"Paths": ["/ipfs", "/ipns"]
}
}'
Backward-compatible: content paths redirect to subdomains:
http://subdomain-gw.example.com/ipfs/{cid} → http://{cid}.ipfs.subdomain-gw.example.com
X-Forwarded-Proto: if you run Kubo behind a reverse proxy that provides TLS, make it add an X-Forwarded-Proto: https header so users are redirected to https://, not http://. It also inlines DNSLink names into a single DNS label, so they work with a wildcard TLS cert (details). The NGINX directive is proxy_set_header X-Forwarded-Proto "https";:
http://subdomain-gw.example.com/ipfs/{cid} → https://{cid}.ipfs.subdomain-gw.example.com
http://subdomain-gw.example.com/ipns/your-dnslink.example.org → https://your--dnslink-example-org.ipns.subdomain-gw.example.com
X-Forwarded-Host: override the gateway host from the request with X-Forwarded-Host: example.net:
http://subdomain-gw.example.com/ipfs/{cid} → http://{cid}.ipfs.example.net
A path gateway
serves content under a path (http://path-gw.example.com/ipfs/{cid}), with no
Origin separation between content roots.
$ ipfs config --json Gateway.PublicGateways '{
"path-gw.example.com": {
"UseSubdomains": false,
"Paths": ["/ipfs", "/ipns"]
}
}'
A DNSLink gateway
resolves the DNSLink name in the Host header of each request. It is on by
default (NoDNSLink: false):
ipfs config --json Gateway.NoDNSLink false
To serve a single DNSLink site and nothing else, combine NoFetch and
NoDNSLink. Disable fetching remote data (NoFetch: true) and DNSLink at
unknown hostnames (NoDNSLink: true), then enable DNSLink for one hostname
whose data is already on the node, without exposing any content-addressing
Paths:
$ ipfs config --json Gateway.NoFetch true
$ ipfs config --json Gateway.NoDNSLink true
$ ipfs config --json Gateway.PublicGateways '{
"dnslink-site.example.com": {
"NoDNSLink": false,
"Paths": []
}
}'
The Gateway.* configuration options are described in the
config documentation. See Gateway recipes
above for common setups.
The gateway's log level can be changed with this command:
> ipfs log level core/server debug
When deploying Kubo's gateway in production, be aware of these important considerations:
<a id="reverse-proxy"></a>
[!IMPORTANT] Reverse Proxy: When running Kubo behind a reverse proxy (such as nginx), the original
Hostheader must be forwarded to Kubo forGateway.PublicGatewaysto work. Kubo uses theHostheader to match configured hostnames and detect subdomain gateway patterns like{cid}.ipfs.example.orgor DNSLink hostnames.If the
Hostheader is not forwarded correctly, Kubo will not recognize the configured gateway hostnames and requests may be handled incorrectly.If
X-Forwarded-Protois not set, redirects over HTTPS will use wrong protocol and DNSLink names will not be inlined for subdomain gateways.Example: minimal nginx configuration for
example.orgnginxserver { listen 80; listen [::]:80; # IMPORTANT: Include wildcard to match subdomain gateway requests. # The dot prefix matches both apex domain and all subdomains. server_name .example.org; location / { proxy_pass http://127.0.0.1:8080; # IMPORTANT: Forward the original Host header to Kubo. # Without this, PublicGateways configuration will not work. proxy_set_header Host $host; # IMPORTANT: X-Forwarded-Proto is required for correct behavior: # - Redirects will use https:// URLs when set to "https" # - DNSLink names will be inlined for subdomain gateways # (e.g., /ipns/en.wikipedia-on-ipfs.org → en-wikipedia--on--ipfs-org.ipns.example.org) proxy_set_header X-Forwarded-Proto $scheme; proxy_set_header X-Forwarded-Host $host; } }Common mistakes to avoid:
Missing wildcard in
server_name: Using onlyserver_name example.org;will not match subdomain requests like{cid}.ipfs.example.org. Always include*.example.orgor use the dot prefix.example.org.Wrong
Hostheader value: Usingproxy_set_header Host $proxy_host;sends the backend's hostname (e.g.,127.0.0.1:8080) instead of the originalHostheader. Always use$hostor$http_host.Missing
Hostheader entirely: Ifproxy_set_header Hostis not specified, nginx defaults to$proxy_host, which breaks gateway routing.
[!IMPORTANT] Timeouts: Configure
Gateway.RetrievalTimeoutto terminate stalled transfers (resets on each data write, catches unresponsive operations), andGateway.MaxRequestDurationas a fallback deadline (default: 1 hour, catches cases when other timeouts are misconfigured or fail to fire).
[!IMPORTANT] Rate Limiting: Use
Gateway.MaxConcurrentRequeststo protect against traffic spikes.
[!IMPORTANT] CDN/Cloudflare: If using Cloudflare or other CDNs with deserialized responses enabled, review
Gateway.MaxRangeRequestFileSizeto avoid excess bandwidth billing from range request bugs. Cloudflare users may need additional protection via Cloudflare Snippets.
For convenience, the gateway (mostly) acts like a normal web-server when serving a directory:
/, append a / and redirect. This applies to
any directory request and helps avoid serving duplicate content from
different paths.<sup>†</sup>index.html file, serve it.<sub><sup>†</sup>This redirect is skipped if the query string contains a
go-get=1 parameter. See PR#3963
for details</sub>
You can use an IPFS gateway to serve static websites at a custom domain using DNSLink. See Example: IPFS Gateway for instructions.
When downloading files, browsers will usually guess a file's filename by looking
at the last component of the path. Unfortunately, when linking directly to a
file (with no containing directory), the final component is just a CID
(bafy.. or Qm...). This isn't exactly user-friendly.
To work around this issue, you can add a filename=some_filename parameter to
your query string to explicitly specify the filename. For example:
http://127.0.0.1:8080/ipfs/QmfM2r8seH2GiRaC4esTjeraXEachRt8ZsSeGaWTPLyMoG?filename=hello_world.txt
When you try to save the page above, your browser will use the passed filename instead of a CID.
It is possible to skip browser rendering of supported filetypes (plain text,
images, audio, video, PDF) and trigger immediate "save as" dialog by appending
&download=true:
An explicit response format can be requested using ?format=raw|car|.. URL parameter,
or by sending Accept: application/vnd.ipld.{format} HTTP header with one of supported content types.
Majority of resources can be retrieved trustlessly by requesting specific content type via Accept header or ?format=raw|car|ipns-record URL query parameter.
See trustless gateway specification and verifiable retrieval documentation for more details.
application/vnd.ipld.rawReturns a byte array for a single raw block.
Sending such requests for /ipfs/{cid} allows for efficient fetch of blocks with data
encoded in custom format, without the need for deserialization and traversal on the gateway.
This is the equivalent of ipfs block get.
application/vnd.ipld.carReturns a CAR stream for a DAG or a subset of it.
The dag-scope parameter controls which blocks are included: all (default, entire DAG),
entity (logical unit like a file), or block (single block). For UnixFS files,
entity-bytes enables byte range requests. See IPIP-402
for details.
This is a rough equivalent of ipfs dag export.
application/vnd.ipfs.ipns-recordOnly works on /ipns/{ipns-name} content paths that use cryptographically signed IPNS Records.
Returns IPNS Record in Protobuf Serialization Format which can be verified on end client, without trusting gateway.