Work in progress

This document is not complete. Please check back soon for updates.

XRPC

XRPC is a general purpose server-to-server messaging protocol. It was created for the AT Protocol but is a generic communications layer which can be applied to multiple use-cases (and which does not include any atproto-specific semantics). The repository data layer and social applications operate as layers atop XRPC.

┌─────────────────────┐
│ Social Applications │  Application semantics
└─┰──────────┰────────┘
  ┃          ┃
  ┃  ┌───────▽────────┐
  ┃  │  Repositories  │  Block & record storage
  ┃  └───────┰────────┘
  ┃          ┃
┌─▽──────────▽────────┐
│        XRPC         │  Wire protocol
└─────────────────────┘

Features:

  • Contract-oriented. All "methods" in XRPC are declared by schemas which define the accepted inputs and outputs. Schemas are globally identified and published as machine-readable documents. This helps ensure correctness and consistency across an open network of services.
  • HTTP-based. XRPC methods are transported using HTTP/S, using GET or POST methods depending on the behavior. This makes XRPC easy to understand and easy to integrate into existing tech stacks.
  • Cacheable. XRPC's "query" methods are designed to cache well with common HTTP-based caching techniques.
  • Support for multiple encodings. XRPC supports structured data (JSON) and unstructured binary blobs.

TODOs

  • Authentication
  • Schema versioning & extensibility
  • Define getSchema

Specification

XRPC supports client-to-server and server-to-server messaging over HTTP/S. Each user has a "Personal Data Server (PDS)" which acts as their agent in the network, meaning most (if not all) of their communication is routed through their PDS.

┌────────┐            ┌────────┐
│ Server │ ◀──XRPC──▶ │ Server │
└────────┘            └────────┘
    ▲
    │
   XRPC
    │
    ▼
┌────────┐
│ Client │
└────────┘

Methods

XRPC "Methods" possess the following attributes:

  • ID: The ID of the schema for the method's inputs and outputs.
  • Type: Query (non-effectful, cacheable) or Procedure (effectful, non-cacheable).
  • Parameters: Encoded in the URI query segment. Affects caching.
  • Input: The request body.
  • Output: The response body.

Calls to a method must specify the ID, Parameters, Input, and certain HTTP Headers for the request. Likewise the return value must provide some information about the HTTP response. Therefore XRPC does not fully abstract away the semantics of HTTP when used in APIs.

Method IDs

Methods are identified using NSIDs, a form of Reverse Domain-Name Notation.

Some example method IDs:

com.example.status()
io.social.getFeed()
net.users.bob.ping()

Method schemas

Method schemas are encoded in JSON using Lexicon Schema Documents.

Schema distribution

Method schemas are designed to be machine-readable and network-accessible. While it is not currently required that a schema is available on the network, it is strongly advised to publish schemas so that a single canonical & authoritative representation is available to consumers of the method.

To fetch a schema, a request must be sent to the builtin getSchema method. This request is sent to the authority of the NSID.

Built-in methods

getSchema

TODO

Requests

HTTP Method

The HTTP Method used depends on the type specified by the method schema.

Type Method
query GET
procedure POST

Path

All requests are sent to the /xrpc/{methodId} path on the target server. For example, a call to the io.social.getFeed method would be sent to /xrpc/io.social.getFeed path.

The parameters (as specified in the Method schema) are encoded as query parameters. The values should be encoded using the following algorithm in pseudo-javascript:

function encodeParam (paramType, value) {
  if (paramType === 'boolean') {
    return value ? 'true' : 'false'
  } else {
    return encodeURIComponent(value)
  }
}

If a default value is specified in the method schema, that value should be included in requests to ensure consistent caching behaviors.

Headers

Header Usage
Content-Type Must specify the encoding of the request body if present.
Authorization May specify the authentication data if present. See Authentication for more information.

Responses

Response types are identified by the HTTP status codes.

200 Request successful

The request has succeeded. Expectations:

  • Content-Type header must be populated.
  • Response body will vary by the method interface.

400 Invalid request

The request is invalid and was not processed.

401 Authentication required

The request cannot be processed without authentication. Expectations:

  • WWW-Authenticate header must be populated with an authentication challenge. See Authentication for more information.

403 Forbidden

The user lacks the needed permissions to access the method.

404 XRPC not supported

The interpretation of a 404 response is somewhat unique for XRPC. A 404 indicates that the server does not provide a resource at the given location (/xrpc) meaning the server does not support XRPC.

To indicate that the given procedure is not implemented, use the 501 response.

413 Payload too large

The payload of the request is larger than the server is willing to process. Payload size-limits are decided by each server.

429 Rate limit exceeded

The client has sent too many requests. Rate-limits are decided by each server. Expectations:

  • Retry-After header may be populated with the amount of time that must pass before the next request.

500 Internal server error

The server reached an unexpected condition during processing.

501 Method not implemented

The server does not implement the requested method.

502 A request to upstream failed

The execution of the procedure depends on a call to another server which has failed.

503 Not enough resources

The server is under heavy load and can't complete the request.

504 A request to upstream timed out

The execution of the procedure depends on a call to another server which timed out.

Remaining codes

Any response code not explicitly enumerated should be handled as follows:

  • 1xx treat as a 404
  • 2xx treat as a 200
  • 3xx treat as a 404 (redirects are not supported)
  • 4xx treat as a 400
  • 5xx treat as a 500

Authentication

TODO

Custom error codes and descriptions

In non-200 (error) responses, services may respond with a JSON body which matches the following schema:

interface XrpcErrorDescription {
  error?: string
  message?: string
}

The error field of the response body should map to an error name defined in the method's Lexicon schema. This enables more specific error-handling by client software. This is especially advised on 400, 500, and 502 responses where further information will be useful.

Join the Bluesky private beta.

The AT Protocol will launch soon.

Join the waitlist