Path Gateway Specification

status: reliable
Marcin Rataj (Protocol Labs) GitHub
Adrian Lanzafame (Protocol Labs) GitHub
Vasco Santos (Protocol Labs) GitHub
Oli Evans (Protocol Labs) GitHub
Henrique Dias (Protocol Labs) GitHub
Commit History
GitHub ipfs/specs (pull requests, new issue, open issues)

The most versatile form of IPFS Gateway is a Path Gateway.

It exposes namespaces like /ipfs/ and /ipns/ under HTTP server root and provides basic primitives for integrating IPFS resources within existing HTTP stack.

Note: additional Web Gateways aimed for website hosting and web browsers extend the below spec and are defined in [subdomain-gateway] and [dnslink-gateway]. There is also a minimal [trustless-gateway] specification for use cases where client prefers to perform all validation locally.


Path Gateway provides HTTP interface for requesting content-addressed data at specified content path.

1.1 GET /ipfs/{cid}[/{path}][?{params}]

Downloads data at specified immutable content path.

1.2 HEAD /ipfs/{cid}[/{path}][?{params}]

Same as GET, but does not return any payload.

Implementations SHOULD limit the scope of IPFS data transfer triggered by HEAD requests to a minimal DAG subset required for producing response headers such as X-Ipfs-Roots, Content-Length and Content-Type.

1.2.1 only-if-cached HEAD behavior

HTTP client can send HEAD request with Cache-Control: only-if-cached to disable IPFS data transfer and inexpensively probe if the gateway has the data cached.

Implementation MUST ensure that handling only-if-cached HEAD response is fast and does not generate any additional I/O such as IPFS data transfer. This allows light clients to probe and prioritize gateways which already have the data.

1.3 GET /ipns/{name}[/{path}][?{params}]

Downloads data at specified mutable content path.

Implementation must resolve the name to a CID, then serve response behind a /ipfs/{resolved-cid}[/{path}][?{params}] content path.

1.4 HEAD /ipns/{name}[/{path}][?{params}]

Same as GET, but does not return any payload.

2. HTTP Request

2.1 Request Headers

All request headers are optional.

2.1.1 If-None-Match (request header)

Used for HTTP caching.

Enables advanced cache control based on Etag, allowing client and server to skip data transfer if previously downloaded payload did not change.

The Gateway MUST compare Etag values sent in If-None-Match with Etag that would be sent with response. Positive match MUST return HTTP status code 304 (Not Modified), without any payload.

2.1.2 Cache-Control (request header)

Used for HTTP caching. only-if-cached

Client can send Cache-Control: only-if-cached to request data only if the gateway already has the data (e.g. in local datastore) and can return it immediately.

If data is not cached locally, and the response requires an expensive remote fetch, a 412 Precondition Failed HTTP status code should be returned by the gateway without any payload or specific HTTP headers.

NOTE: when processing a request for a DAG, traversing it and checking every CID might be too expensive. Implementations SHOULD implement own heuristics to maximize cache hits while minimizing performance cost of checking if the entire DAG is locally cached. A good rule of thumb is to at the minimum test if the root block is in the local cache.

2.1.3 Accept (request header)

Can be used for requesting specific response format

For example:

2.1.4 Range (request header)

Range can be used for requesting specific byte range of UnixFS files and raw blocks.

Gateway implementations SHOULD be smart enough to require only the minimal DAG subset necessary for handling the range request.

Gateways SHOULD support single range requests. The support of more than one range is optional: implementation MAY decide to not support more than one range.

For more advanced use cases such as partial DAG/CAR streaming, or non-UnixFS data structures, see dag-scope and entity-bytes from [ipip-0402] and ordered CARs from [ipip-0412].

2.1.5 Service-Worker (request header)

Mentioned here for security reasons and should be implemented with care.

This header is sent by web browser attempting to register a service worker script for a specific scope. Allowing too broad scope can allow a single content root to take control over gateway endpoint. It is important for implementations to handle this correctly.

Service Worker should only be allowed under specific to content roots under /ipfs/{cid}/ and /ipns/{name}/ (IMPORTANT: note the trailing slash).

Gateway should refuse attempts to register a service worker for entire /ipfs/cid or /ipns/name (IMPORTANT: when trailing slash is missing).

Requests to these paths with Service-Worker: script MUST be denied by returning HTTP 400 Bad Request error.

2.2 Request Query Parameters

All query parameters are optional.

2.2.1 filename (request query parameter)

Optional, can be used for overriding the filename.

When set, gateway will include it in Content-Disposition header and may use it for Content-Type calculation.


2.2.2 download (request query parameter)

Optional, can be used to request specific Content-Disposition to be set on the response.

Response to HTTP request with download=true MUST include Content-Disposition: attachment[;filename=...] to indicate that client should not render the response.

The attachment context will force user agents such as web browsers to present a 'Save as' dialog instead (prefilled with the value of the filename parameter, if present)

2.2.3 format (request query parameter)

Optional, format=<format> can be used to request specific response format.

This is a URL-friendly alternative to sending an Accept header. These are the equivalents:

  • format=rawAccept: application/vnd.ipld.raw
  • format=carAccept: application/
  • format=tarAccept: application/x-tar
  • format=dag-jsonAccept: application/vnd.ipld.dag-json
  • format=dag-cborAccept: application/vnd.ipld.dag-cbor
  • format=jsonAccept: application/json
  • format=cborAccept: application/cbor
  • format=ipns-recordAccept: application/vnd.ipfs.ipns-record

When both Accept HTTP header and format query parameter are present, Accept SHOULD take precedence.

A Client SHOULD include the format query parameter in the request URL, in addition to the Accept header. This provides the best interoperability and ensures consistent HTTP cache behavior across various gateway implementations.

A Gateway SHOULD include the Content-Location header in the response when:

  • the request contains an Accept header specifying a well-known response format, but the URL does not include the format query parameter
  • the format parameter is present, but does not match the format from Accept

2.2.4 dag-scope (request query parameter)

Optional, can be used to limit the scope of verifiable DAG requests such as CAR, same as dag-scope from [trustless-gateway].

2.2.5 entity-bytes (request query parameter)

Optional, can be used to limit the scope of verifiable DAG requests such as CAR, same as entity-bytes from [trustless-gateway].

2.2.6 car-version (request query parameter)

Optional, specific to CAR requests, same as car-version from [trustless-gateway].

2.2.7 car-order (request query parameter)

Optional, specific to CAR requests, same as car-order from [trustless-gateway].

2.2.8 car-dups (request query parameter)

Optional, specific to CAR requests, same as car-dups from [trustless-gateway].

3. HTTP Response

3.1 Response Status Codes

3.1.1 200 OK

The request succeeded.

If the HTTP method was GET, then data is transmitted in the message body.

3.1.2 206 Partial Content

Partial Content: range request succeeded.

Returned when requested range of data described by Range header of the request.

3.1.3 301 Moved Permanently

Indicates permanent redirection.

The new, canonical URL is returned in the Location header.

3.1.4 400 Bad Request

A generic client error returned when it is not possible to return a better one

3.1.5 404 Not Found

Error to indicate that request was formally correct, but traversal of the requested content path was not possible due to a invalid or missing DAG node.

3.1.6 410 Gone

Error to indicate that request was formally correct, but this specific Gateway refuses to return requested data.

Particularly useful for implementing deny lists, in order to not serve malicious content. The name of deny list and unique identifier of blocked entries can be provided in the response body.

See: Denylists

3.1.7 412 Precondition Failed

Error to indicate that request was formally correct, but Gateway is unable to return requested data under the additional (usually cache-related) conditions sent by the client. Use with only-if-cached
  • Client sends a request with Cache-Control: only-if-cached
  • Gateway does not have requested CIDs in local datastore, and is unable to fetch them from other peers due to only-if-cached condition
  • Gateway returns status code 412 to the client
    • The code 412 is used instead of 504 because only-if-cached is handled by the gateway itself, moving the error to client error range and avoiding confusing server errors in places like the browser console.

3.1.8 429 Too Many Requests

Error to indicate the client has sent too many requests in a given amount of time.

This error response SHOULD include Retry-After HTTP header to indicate how long the client should wait before making a follow-up request.

3.1.10 500 Internal Server Error

A generic server error returned when it is not possible to return a better one.

3.1.11 502 Bad Gateway

Returned immediately when Gateway was not able to produce response for a known reason. For example, when gateway failed to find any providers for requested data.

This error response SHOULD include Retry-After HTTP header to indicate how long the client should wait before retrying.

3.1.12 504 Gateway Timeout

Returned when Gateway was not able to produce response under set time limits. For example, when gateway failed to retrieve data from a remote provider.

There is no generic timeout, Gateway implementations SHOULD set timeouts based on specific use cases.

This error response SHOULD include Retry-After HTTP header to indicate how long the client should wait before retrying.

3.2 Response Headers

3.2.1 Etag (response header)

Used for HTTP caching.

An opaque identifier for a specific version of the returned payload. The unique value must be wrapped by double quotes as noted in Section 8.8.3 of [rfc9110].

In many cases it is not enough to base Etag value on requested CID.

To ensure Etag is unique enough to avoid issues with caching reverse proxies and CDNs, implementations should base it on both CID and response type:

  • By default, etag should be based on requested CID. Example: Etag: "bafy…foo"

  • If a custom format was requested (such as a raw block, CAR), the returned etag should be modified to include it. It could be a suffix.

    • Example: Etag: "bafy…foo.raw"
  • If HTML directory index was generated by the gateway, the etag returned with HTTP response should be based on the version of gateway implementation. This is to ensure proper cache busting if code responsible for HTML generation changes in the future.

    • Example: Etag: "DirIndex-2B423AF_CID-bafy…foo"
  • When a gateway can’t guarantee byte-for-byte identical responses, a “weak” etag should be used.

    • Example: If CAR is streamed, and blocks arrive in non-deterministic order, the response should have Etag: W/"bafy…".
    • Example: If TAR stream is generated by traversing an UnixFS directory in non-deterministic order, the response should have Etag: W/"bafy…foo.x-tar".
  • When responding to Range request, a strong Etag should be based on requested range in addition to CID and response format: Etag: "

3.2.2 Cache-Control (response header)

Used for HTTP caching.

An explicit caching directive for the returned response. Informs HTTP client and intermediate middleware caches such as CDNs if the response can be stored in caches.

Returned directive depends on requested content path and format:

  • Cache-Control: public, max-age=29030400, immutable MUST be returned for every immutable resource under /ipfs/ namespace.

  • Cache-Control: public, max-age=<ttl> SHOULD be returned for mutable resources under /ipns/{id-with-ttl}/ namespace; max-age=<ttl> SHOULD indicate remaining TTL of the mutable pointer such as [ipns-record] or DNSLink TXT record.

    • Implementations MAY place an upper bound on any TTL received, as noted in Section 8 of [rfc2181].
    • If TTL value is unknown, implementations SHOULD not send a Cache-Control
    • No matter if TTL value is known or not, implementations SHOULD always send a Last-Modified header with the timestamp of the record resolution.

3.2.3 Last-Modified (response header)

Optional, used as additional hint for HTTP caching.

Returning this header depends on the information available:

  • The header can be returned with /ipns/ responses when the gateway implementation knows the exact time a mutable pointer was updated by the publisher.

  • When only TTL is known, Cache-Control should be used instead.

  • Legacy implementations set this header to the current timestamp when reading TTL on /ipns/ content paths was not available. This hint was used by web browsers in a process called "Calculating Heuristic Freshness" (Section 4.2.2 of [rfc9111]). Each browser uses different heuristic, making this an inferior, non-deterministic caching strategy.

  • New implementations should not return this header if TTL is not known; providing a static expiration window in Cache-Control is easier to reason about than cache expiration based on the fuzzy “heuristic freshness”.

3.2.4 Content-Type (response header)

Returned with custom response formats such as application/ or application/vnd.ipld.raw. CAR must be returned with explicit version. Example: Content-Type: application/; version=1

When deserialized responses are enabled, and no explicit response format is provided with the request, and the requested data itself has no built-in content type metadata, implementations SHOULD perform content type sniffing based on file name (from url path, or optional filename parameter) and magic bytes to improve the utility of produced responses.

For example:

  • detect plain text file and return Content-Type: text/plain instead of application/octet-stream
  • detect SVG image and return Content-Type: image/svg+xml instead of text/xml

3.2.5 Content-Disposition (response header)

Returned when download, filename query parameter, or a custom response format such as car or raw block are used.

The first parameter passed in this header indicates if content should be displayed inline by the browser, or sent as an attachment that opens the “Save As” dialog:

  • Content-Disposition: inline is the default, returned when request was made with download=false or a custom filename was provided with the request without any explicit download parameter.
  • Content-Disposition: attachment is returned only when request was made with the explicit download=true

The remainder is an optional filename parameter that will be prefilled in the “Save As” dialog.

NOTE: when the filename includes non-ASCII characters, the header must include both ASCII and UTF-8 representations for compatibility with legacy user agents and existing web browsers.

To illustrate, ?filename=testтест.pdf should produce: Content-Disposition inline; filename="test____.jpg"; filename*=UTF-8''test%D1%82%D0%B5%D1%81%D1%82.jpg

  • ASCII representation must have non-ASCII characters replaced with _
  • UTF-8 representation must be wrapped in Percent Encoding (Section 2.1 of [rfc3986]).
    • NOTE: UTF-8'' is not a typo – see Section 3.2.3 of [rfc8187].

Content-Disposition must be also set when a binary response format was requested:

  • Content-Disposition: attachment; filename="<cid>.car" should be returned with Content-Type: application/ responses to ensure client does not attempt to render streamed bytes. CID and .car file extension should be used if a custom filename was not provided with the request.

  • Content-Disposition: attachment; filename="<cid>.bin" should be returned with Content-Type: application/vnd.ipld.raw responses to ensure client does not attempt to render raw bytes. CID and .bin file extension should be used if a custom filename was not provided with the request.

3.2.6 Content-Location (response header)

Returned when a non-default content format has been negotiated with the Accept header but format was missing from the URL.

The value of this field SHOULD include the URL of the resource with the format query parameter included, so that generic HTTP caches can store deserialized, CAR, and block responses separately.

For example, a request to /ipfs/{cid} with Accept: application/vnd.ipld.raw SHOULD return a Content-Location: /ipfs/{cid}?format=raw header in order for block response to be cached separately from deserialized one.

3.2.7 Content-Length (response header)

Represents the length of returned HTTP payload.

NOTE: the value may differ from the real size of requested data if compression or chunked Transfer-Encoding are used. See ipfs/specs#461.

3.2.8 Content-Range (response header)

Returned only when request was a Range request.

See Section 14.4 of [rfc9110].

3.2.9 Accept-Ranges (response header)

Optional, returned to explicitly indicate if gateway supports partial HTTP Range requests for a specific resource.

For example, Accept-Ranges: none should be returned with application/ responses if the block order in CAR stream is not deterministic.

3.2.10 Location (response header)

Returned only when response status code is 301 Moved Permanently. The value informs the HTTP client about new URL for requested resource. Use in directory URL normalization

Gateway MUST return a redirect when a valid UnixFS directory was requested without the trailing /, for example:

  • response for UnixFS directory at (no trailing slash) MUST be HTTP 301 redirect with Location: /ipns/

This normalization is required for directory-based relative links and security scopes such as Service Worker registrations to work correctly. It also ensures the same behavior on path gateways ( with trailing /) and origin-isolated HTTP contexts or non-HTTP URLs like ipfs://cid, where empty path component is implicit /. Use in interop with Subdomain Gateway

See Location section of [subdomain-gateway].

3.2.11 X-Ipfs-Path (response header)

Used for HTTP caching and indicating the IPFS address of the data.

Indicates the original, requested content path before any path resolution and traversal is performed.

Example: X-Ipfs-Path: /ipns/k2..ul6/subdir/file.txt

3.2.12 X-Ipfs-Roots (response header)

Used for HTTP caching.

A way to indicate all CIDs required for resolving logical roots (path segments) from X-Ipfs-Path. The main purpose of this header is allowing HTTP caches to make smarter decisions about cache invalidation.

Below, an example to illustrate how X-Ipfs-Roots is constructed from X-Ipfs-Path pointing at a DNSLink.

The traversal of /ipns/ includes a HAMT-sharded UnixFS directory /wiki/.

This header only cares about logical roots (one per URL path segment):

  1. /ipns/en.wikipedia-on-ipfs.orgbafybeiaysi4s6lnjev27ln5icwm6tueaw2vdykrtjkwiphwekaywqhcjze
  2. /ipns/
  3. /ipns/

Final array of roots:

X-Ipfs-Roots: bafybeiaysi4s6lnjev27ln5icwm6tueaw2vdykrtjkwiphwekaywqhcjze,bafybeihn2f7lhumh4grizksi2fl233cyszqadkn424ptjajfenykpsaiw4,bafkreibn6euazfvoghepcm4efzqx5l3hieof2frhp254hio5y7n3hv5rma

NOTE: while the first CID will change every time any article is changed, the last root (responsible for specific article or a subdirectory) may not change at all, allowing for smarter caching beyond what standard Etag offers.

3.2.13 X-Content-Type-Options (response header)

Optional, present in certain response types:

  • X-Content-Type-Options: nosniff should be returned with application/ and application/vnd.ipld.raw responses to indicate that the Content-Type should be followed and not be changed. This is a security feature, ensures that non-executable binary response types are not used in <script> and <style> HTML tags.

3.2.14 Retry-After (response header)

Gateway returns this header with error responses such as 429 Too Many Requests or 504 Gateway Timeout.

The "Retry-After" header indicates how long the user agent ought to wait before making a follow-up request.

See Section 10.2.3 of [rfc9110].

3.2.15 Server-Timing (response header)

Optional. Implementations MAY use this header to communicate one or more metrics and descriptions for the given request-response cycle.

See Server-Timing at W3C: Server Timing.

3.2.16 Traceparent (response header)

Optional. Implementations MAY use this header to return a globally unique identifier to help in debugging errors and performance issues.

See Traceparent at W3C: Trace Context.

3.2.17 Tracestate (response header)

Optional. Implementations MAY use this header to return a additional vendor-specific trace identification information across different distributed tracing systems and is a companion header for the Traceparent header.

See Tracestate at W3C: Trace Context.

3.3 Response Payload

Data sent with HTTP response depends on the type of the requested IPFS resource, and the requested response type.

By default, implicit deserialized response type is based on Accept header and the codec of the resolved CID:

The following response types require an explicit opt-in, can only be requested with format query parameter or Accept header:

4. Appendix: notes for implementers

4.1 Content resolution

Content resolution is a process of turning an HTTP request into an IPFS content path, and then traversing it until the content identifier (CID) is found.

4.1.1 Finding the content root

Path Gateway decides what content to serve by taking the path from the URL requested and splitting it into two parts: the CID and the remainder of the path.

The CID provides the starting point, often called content root. The remainder of the path, if present, will be used as instructions to traverse IPLD data, starting from that data which the CID identified.

Note: Other types of gateway may allow for passing CID by other means, such as Host header, changing the rules behind path splitting. (See and

4.1.2 Traversing remaining path

After the content root CID is found, the remaining of the path should be traversed and resolved. Depending on the data type, that may occur through UnixFS pathing, or DAG-JSON, and DAG-CBOR pathing.

4.1.3 Traversing through UnixFS

UnixFS is an abstraction over the low level logical DAG-PB pathing from IPLD, providing a better user experience:

  • Example of UnixFS pathing: /ipfs/cid/dir-name/file-name.txt

For more details regarding DAG-PB pathing, please read the "Path Resolution" section of this document.

4.1.4 Traversing through DAG-JSON and DAG-CBOR

Traversing through DAG-JSON and DAG-CBOR is possible through fields that encode a link:

  • DAG-JSON: link are represented as a base encoded CID under the / reserved namespace, see specification.
  • DAG-CBOR: links are tagged with CBOR tag 42, indicating that they encode a CID, see specification.

Note: pathing into IPLD Kind other than Link (CID) is not supported at the moment. Implementations should return HTTP 501 Not Implemented when fully resolved content path has any remainder left. This feature may be specified in a future IPIP that introduces data onboarding and IPLD Patch semantics.

4.1.5 Handling traversal errors

Gateway MUST respond with HTTP error when it is not possible to traverse the requested content path:

  • 404 Not Found should be returned when the root CID is valid and traversable, but the DAG it represents does not include content path remainder.
    • Error response body should indicate which part of immutable content path (/ipfs/{cid}/path/to/file) is missing
  • 400 Bad Request should be returned when the root CID under the ipfs namespace is invalid.
  • 500 Internal Server Error can be used for remaining traversal errors, such as domains that cannot be resolved, or IPNS keys that cannot be resolved.

4.2 Best practices for HTTP caching

4.3 Denylists

Optional, but encouraged.

Implementations are encouraged to support pluggable denylists to allow IPFS node operators to opt into not hosting previously flagged content.

Gateway MUST respond with HTTP error when requested CID is on any of active denylists:

Gateway implementation MAY apply some denylists by default as long the gateway operator is able to inspect and modify the list of denylists that are applied.

Examples of public deny lists

4.4 Generated HTML with directory index

While implementations decide on the way HTML directory listing is generated and presented to the user, following below suggestions is advised.

Linking to alternative response types such as CAR and dag-json allows clients to consume directory listings programmatically without the need for parsing HTML.

Directory index response time should not grow with the number of items in a directory. It should be always fast, even when a directory has 10k of items.

The usual optimizations involve:

A. References

DNSLink Gateway Specification. Marcin Rataj; Thibault Meunier. 2022-11-09. URL:
IPIP-0288: TAR Response Format on HTTP Gateways. Henrique Dias; Marcin Rataj. 2022-06-10. URL:
IPIP-0402: Partial CAR Support on Trustless Gateways. Hannah Howard; Adin Schmahmann; Rod Vagg; Marcin Rataj. 2023-04-17. URL:
IPIP-0412: Signaling Block Order in CARs on HTTP Gateways. Marcin Rataj; Jorropo. 2023-05-15. URL:
IPNS Record and Protocol. Vasco Santos; Steve Allen; Marcin Rataj; Henrique Dias; Gus Eggert. 2023-10-03. URL:
Key words for use in RFCs to Indicate Requirement Levels. S. Bradner. IETF. March 1997. Best Current Practice. URL:
Clarifications to the DNS Specification. R. Elz; R. Bush. IETF. July 1997. Proposed Standard. URL:
Uniform Resource Identifier (URI): Generic Syntax. T. Berners-Lee; R. Fielding; L. Masinter. IETF. January 2005. Internet Standard. URL:
An HTTP Status Code to Report Legal Obstacles. T. Bray. IETF. February 2016. Proposed Standard. URL:
Indicating Character Encoding and Language for HTTP Header Field Parameters. J. Reschke. IETF. September 2017. Proposed Standard. URL:
HTTP Semantics. R. Fielding, Ed.; M. Nottingham, Ed.; J. Reschke, Ed.. IETF. June 2022. Internet Standard. URL:
HTTP Caching. R. Fielding, Ed.; M. Nottingham, Ed.; J. Reschke, Ed.. IETF. June 2022. Internet Standard. URL:
Subdomain Gateway Specification. Marcin Rataj; Adrian Lanzafame; Vasco Santos; Oli Evans; Thibault Meunier; Steve Loeppky. 2023-01-28. URL:
Trustless Gateway Specification. Marcin Rataj; Henrique Dias. 2023-06-20. URL:
  1. No definition for term "car-version".
  2. No definition for term "car-order".
  3. No definition for term "car-dups".