Skip to content

reactiveui/refit

Repository files navigation

Refit

Refit: The automatic type-safe REST library for modern .NET

Build codecov

Refit Refit.HttpClientFactory Refit.Newtonsoft.Json Refit.Testing
NuGet NuGet NuGet NuGet NuGet

Refit is a library heavily inspired by Square's Retrofit library, and it turns your REST API into a live interface:

public interface IGitHubApi
{
    [Get("/users/{user}")]
    Task<User> GetUser(string user);
}

The RestService class generates an implementation of IGitHubApi that uses HttpClient to make its calls:

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com");
var octocat = await gitHubApi.GetUser("octocat");

.NET supports registering Refit clients via HttpClientFactory:

services
    .AddRefitClient<IGitHubApi>()
    .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.github.com"));

To test the clients you build with Refit, the Refit.Testing package lets you stub responses and verify requests with a declarative route table — see Testing your Refit clients.

Table of Contents

Sponsors

Refit is sponsored by the following:

lombiq logojetbrains logoclaude logo

Where does this work?

Refit currently supports the following platforms and modern .NET targets:

  • WinUI
  • Desktop .NET Framework 4.6.2+
  • .NET 8 / 9 / 10 / 11
  • Blazor
  • Uno Platform

SDK Requirements

Breaking changes and release notes

Breaking changes and the notable additions for each major version — including the V14 move of the reflection request builder into the opt-in Refit.Reflection package — are documented in Breaking changes and release notes.

Source generation

The Refit package ships Roslyn source generators. A PackageReference to Refit gets you generated clients at build time — no extra package.

The generated code targets C# 7.3. On C# 8 or newer, the generator also emits nullable directives and annotations.

You create generated clients with the normal APIs:

var api = RestService.For<IGitHubApi>("https://api.github.com");

or through Refit.HttpClientFactory:

services
    .AddRefitClient<IGitHubApi>()
    .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.github.com"));

Generated clients use the same RefitSettings you pass to RestService.For<T> or AddRefitClient<T>, and honor settings such as:

  • ContentSerializer
  • UrlParameterFormatter
  • UrlParameterKeyFormatter
  • CollectionFormat
  • AuthorizationHeaderValueGetter
  • ExceptionFactory
  • DeserializationExceptionFactory
  • TransportExceptionFactory
  • HttpRequestMessageOptions
  • Version and VersionPolicy

Automatic client registration

On .NET 5 and newer the generator emits a module initializer that registers every generated client factory at assembly load. RestService.ForGenerated<T> and AddRefitGeneratedClient<T> then resolve the client with no runtime reflection. RestService.For<T> skips the reflection request builder for fully generated interfaces. That keeps trimmed and Native AOT apps reflection-free.

.NET Framework (net462–net481) gets no automatic registration. ModuleInitializerAttribute arrived in .NET 5 and isn't in the .NET Framework BCL, so Refit emits the initializer only for .NET 5+ targets. Raising a project's <LangVersion> to 9 doesn't change that — the missing type is the blocker, not the language version. (It does switch on modern generated syntax like nullable annotations.)

The generated inline code still runs. But RestService.For<T> finds the client by runtime type lookup and builds the reflection request builder, so you must reference the opt-in Refit.Reflection package. .NET Framework has no trimming or AOT, so this costs nothing there.

ForGenerated<T> and AddRefitGeneratedClient<T> need that registration. On .NET Framework they throw unless you register the factory yourself with RegisterGeneratedFactory<T> / RegisterGeneratedSettingsFactory<T> at startup.

Generated-only client creation

For Native AoT or trimmed apps, RestService.ForGenerated<T> creates a client only when the generator registered an implementation for that interface:

using var client = new HttpClient
{
    BaseAddress = new Uri("https://api.github.com")
};

var api = RestService.ForGenerated<IGitHubApi>(client);

ForGenerated<T> never falls back to the reflection client. When the generated client builds every request directly, Refit skips the reflection request builder too. If no generated implementation is registered, it throws an InvalidOperationException pointing back to source generation setup.

Generated request building

By default the generator builds requests directly. Instead of a method body that calls the reflective pipeline through BuildRestResultFuncForMethod, the generated client creates the HttpRequestMessage, applies headers, properties, and body, then dispatches it through Refit's runtime helpers.

This cuts runtime reflection, metadata lookup, argument boxing, and delegate construction. It covers most request shapes:

  • parameters that appear in the path template
  • query parameters: auto-appended, [AliasAs], [Query] (including Format and CollectionFormat), scalar collections, [QueryName] flags and [Encoded] values
  • implicit [Body] detection on POST/PUT/PATCH
  • static [Headers]
  • dynamic [Header] parameters
  • [HeaderCollection] dictionaries
  • [Body] content
  • [Multipart] form uploads whose parts are StreamPart/ByteArrayPart/FileInfoPart (or a MultipartItem subclass), Stream, string, FileInfo, byte[], HttpContent, a date/time or Guid value, or an enumerable of these (an object part serialized through the content serializer still falls back to the runtime builder)
  • [Property] parameters
  • [Property] interface properties
  • cancellation tokens
  • Task, Task<T>, Task<ApiResponse<T>>, and related response wrappers
  • IAsyncEnumerable<T> for streamed responses

For a shape the generator can't emit yet, that method falls back to the runtime request builder.

Turn generated request building off in a project file:

<PropertyGroup>
  <RefitGeneratedRequestBuilding>false</RefitGeneratedRequestBuilding>
</PropertyGroup>

That keeps the generated interface implementation but routes its methods through the reflective request builder.

Disable source generation entirely:

<PropertyGroup>
  <DisableRefitSourceGenerator>true</DisableRefitSourceGenerator>
</PropertyGroup>

Most applications should leave both settings unset.

Analyzer diagnostics and code fixes

The main package also ships analyzers. They flag common interface issues at compile time:

  • methods or properties on Refit interfaces that cannot be generated or called by Refit
  • route templates that use backslashes instead of forward slashes
  • methods with more than one CancellationToken parameter
  • [HeaderCollection] parameters that are not IDictionary<string, string>
  • methods with more than one [Body] parameter
  • [Multipart] methods that also declare a [Body] parameter

Mechanical fixes ship as code fixes: replacing route backslashes with forward slashes, and changing an invalid [HeaderCollection] parameter to IDictionary<string, string>.

API Attributes

Every method must have an HTTP attribute that provides the request method and relative URL. There are six built-in annotations: Get, Post, Put, Delete, Patch and Head. The relative URL of the resource is specified in the annotation.

[Get("/users/list")]

You can also specify query parameters in the URL:

[Get("/users/list?sort=desc")]

A request URL can be updated dynamically using replacement blocks and parameters on the method. A replacement block is an alphanumeric string surrounded by { and }.

If the name of your parameter doesn't match the name in the URL path, use the AliasAs attribute.

[Get("/group/{id}/users")]
Task<List<User>> GroupList([AliasAs("id")] int groupId);

A request url can also bind replacement blocks to a custom object

[Get("/group/{request.groupId}/users/{request.userId}")]
Task<List<User>> GroupList(UserGroupRequest request);

class UserGroupRequest{
    int groupId { get;set; }
    int userId { get;set; }
}

When the bound object is a generic method's type parameter, the placeholders are resolved against a class constraint at compile time, so a constrained generic method is generated inline (no reflection fallback, no RF006):

// Generated inline: {request.groupId}/{request.userId} bind against UserGroupRequest.
[Get("/group/{request.groupId}/users/{request.userId}")]
Task<List<User>> GroupList<T>(T request) where T : UserGroupRequest;

An unconstrained generic parameter (GroupList<T>(T request)) still falls back to the reflection request builder, because the concrete type - and its bound properties - are only known at call time.

Parameters that are not specified as a URL substitution will automatically be used as query parameters. This is different than Retrofit, where all parameters must be explicitly specified.

The comparison between parameter name and URL parameter is not case-sensitive, so it will work correctly if you name your parameter groupId in the path /group/{groupid}/show for example.

[Get("/group/{groupid}/users")]
Task<List<User>> GroupList(int groupId, [AliasAs("sort")] string sortOrder);

GroupList(4, "desc");
>>> "/group/4/users?sort=desc"

Round-tripping route parameter syntax: Forward slashes aren't encoded when using a double-asterisk (**) catch-all parameter syntax.

During link generation, the routing system encodes the value captured in a double-asterisk (**) catch-all parameter ( for example, {**myparametername}) except the forward slashes.

The type of round-tripping route parameter must be string.

[Get("/search/{**page}")]
Task<List<Page>> Search(string page);

Search("admin/products");
>>> "/search/admin/products"

By default Refit throws if a route template contains a placeholder with no matching method argument. If you want to resolve a placeholder later yourself (for example an API-versioning token rewritten inside a DelegatingHandler), set AllowUnmatchedRouteParameters on RefitSettings. The unmatched {token} is then left in the URL verbatim instead of throwing.

var settings = new RefitSettings { AllowUnmatchedRouteParameters = true };
var api = RestService.For<IVersionedApi>("https://api.example.com", settings);

// [Get("/api/{version:apiVersion}/values")]
// the {version:apiVersion} token is left in the path for a DelegatingHandler to replace.

Base address and URL resolution

By default Refit requires relative paths to start with /, prepends the base address path itself, and trims a trailing slash from the base address. If you would rather have the base address and relative URL combined the same way HttpClient and System.Uri do (RFC 3986), set UrlResolution on RefitSettings:

var settings = new RefitSettings { UrlResolution = UrlResolutionMode.Rfc3986 };
var api = RestService.For<IMyApi>("https://api.example.com/api/v1/", settings);

Under UrlResolutionMode.Rfc3986 the leading-slash requirement is relaxed and the trailing slash on the base address is significant, exactly as with HttpClient:

// base address "https://api.example.com/api/v1/"
[Get("values")]   // -> https://api.example.com/api/v1/values  (appended)
[Get("/values")]  // -> https://api.example.com/values         (leading slash replaces the base path)

Note: with generated request building (the default), a leading-slash-less route under the default legacy resolution is validated when the request is built — so the ArgumentException surfaces on the first call rather than from RestService.For<T>(...). Under UrlResolutionMode.Rfc3986 the route is valid and no exception is raised.

Querystrings

Dynamic Querystring Parameters

If you specify an object as a query parameter, all public properties which are not null are used as query parameters. This previously only applied to GET requests, but has now been expanded to all HTTP request methods, partly thanks to Twitter's hybrid API that insists on non-GET requests with querystring parameters. Use the Query attribute to change the behavior to 'flatten' your query parameter object. If using this Attribute you can specify values for the Delimiter and the Prefix which are used to 'flatten' the object.

public class MyQueryParams
{
    [AliasAs("order")]
    public string SortOrder { get; set; }

    public int Limit { get; set; }

    public KindOptions Kind { get; set; }
}

public enum KindOptions
{
    Foo,

    [EnumMember(Value = "bar")]
    Bar
}


[Get("/group/{id}/users")]
Task<List<User>> GroupList([AliasAs("id")] int groupId, MyQueryParams params);

[Get("/group/{id}/users")]
Task<List<User>> GroupListWithAttribute([AliasAs("id")] int groupId, [Query(".","search")] MyQueryParams params);


params.SortOrder = "desc";
params.Limit = 10;
params.Kind = KindOptions.Bar;

GroupList(4, params)
>>> "/group/4/users?order=desc&Limit=10&Kind=bar"

GroupListWithAttribute(4, params)
>>> "/group/4/users?search.order=desc&search.Limit=10&search.Kind=bar"

A similar behavior exists if using a Dictionary, but without the advantages of the AliasAs attributes and of course no intellisense and/or type safety.

You can also specify querystring parameters with [Query] and have them flattened in non-GET requests, similar to:

[Post("/statuses/update.json")]
Task<Tweet> PostTweet([Query]TweetParams params);

Where TweetParams is a POCO, and properties will also support [AliasAs] attributes.

If you need to keep internal-only properties on your query DTO, mark them with one of the standard ignore attributes and Refit will skip them when building the query string:

  • [IgnoreDataMember]
  • [System.Text.Json.Serialization.JsonIgnore]
  • [Newtonsoft.Json.JsonIgnore]

Collections as Querystring parameters

Use the Query attribute to specify format in which collections should be formatted in query string

[Get("/users/list")]
Task Search([Query(CollectionFormat.Multi)]int[] ages);

Search(new [] {10, 20, 30})
>>> "/users/list?ages=10&ages=20&ages=30"

[Get("/users/list")]
Task Search([Query(CollectionFormat.Csv)]int[] ages);

Search(new [] {10, 20, 30})
>>> "/users/list?ages=10%2C20%2C30"

You can also specify collection format in RefitSettings, that will be used by default, unless explicitly defined in Query attribute.

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com",
    new RefitSettings {
        CollectionFormat = CollectionFormat.Multi
    });

Unescape Querystring parameters

Use the QueryUriFormat attribute to specify if the query parameters should be url escaped

[Get("/query")]
[QueryUriFormat(UriFormat.Unescaped)]
Task Query(string q);

Query("Select+Id,Name+From+Account")
>>> "/query?q=Select+Id,Name+From+Account"

For a single pre-encoded value, mark the parameter with [Encoded] (the equivalent of Retrofit's encoded = true) and Refit passes it through verbatim while the rest of the request encodes normally. It applies to query values and to path segments, including round-tripping {**param} segments — the caller becomes responsible for producing valid encoded output:

[Get("/calendars/{calId}/events/{**eventId}")]
Task<CalendarEvent> GetEvent(string calId, [Encoded] string eventId);

GetEvent("work", "3bf0000488fda0ec154ee%40zoho.com")
>>> "/calendars/work/events/3bf0000488fda0ec154ee%40zoho.com"

[Encoded] (and [QueryName] below) are handled by generated request building only; using them on a method that cannot generate inline is a compile-time error (RF007).

Valueless query flags

Some APIs use bare presence-style query switches with no value. Mark a parameter with [QueryName] (the equivalent of Retrofit's @QueryName) and its value becomes the query segment itself; collections render one flag per element and null values are omitted:

[Get("/items")]
Task<List<Item>> List([QueryName] string flag);

List("archived")
>>> "/items?archived"

[Get("/items")]
Task<List<Item>> List([QueryName] string[] flags);

List(["a", "b", "c"])
>>> "/items?a&b&c"

Custom Querystring parameter formatting

Formatting Keys

To customize the format of query keys, you have two main options:

  1. Using the AliasAs Attribute:

    You can use the AliasAs attribute to specify a custom key name for a property. This attribute will always take precedence over any key formatter you specify.

    public class MyQueryParams
    {
        [AliasAs("order")]
        public string SortOrder { get; set; }
    
        public int Limit { get; set; }
    }
    
    [Get("/group/{id}/users")]
    Task<List<User>> GroupList([AliasAs("id")] int groupId, [Query] MyQueryParams params);
    
    params.SortOrder = "desc";
    params.Limit = 10;
    
    GroupList(1, params);

    This will generate the following request:

    /group/1/users?order=desc&Limit=10
    
  2. Using the RefitSettings.UrlParameterKeyFormatter Property:

    By default, Refit uses the property name as the query key without any additional formatting. If you want to apply a custom format across all your query keys, you can use the UrlParameterKeyFormatter property. Remember that if a property has an AliasAs attribute, it will be used regardless of the formatter.

    The following example uses the built-in CamelCaseUrlParameterKeyFormatter:

    public class MyQueryParams
    {
        public string SortOrder { get; set; }
    
        [AliasAs("queryLimit")]
        public int Limit { get; set; }
    }
    
    [Get("/group/users")]
    Task<List<User>> GroupList([Query] MyQueryParams params);
    
    params.SortOrder = "desc";
    params.Limit = 10;

    The request will look like:

    /group/users?sortOrder=desc&queryLimit=10
    

Note: The AliasAs attribute always takes the top priority. If both the attribute and a custom key formatter are present, the AliasAs attribute's value will be used.

Built-in key formatters and naming-convention presets:

Refit ships CamelCaseUrlParameterKeyFormatter, SnakeCaseUrlParameterKeyFormatter, and KebabCaseUrlParameterKeyFormatter. To apply a single naming convention consistently across query keys, form field names and the JSON request body, use the RefitSettings presets — each wires up the matching UrlParameterKeyFormatter and JsonNamingPolicy together:

var api = RestService.For<IMyApi>("https://api.example.com", RefitSettings.SnakeCase());
// also available: RefitSettings.KebabCase() and RefitSettings.CamelCase()

Per-property key prefix and delimiter:

A [Query] attribute on a property of a complex query object customizes that property's key as {prefix}{delimiter}{name} (matching how form fields are named):

public class Form
{
    [Query("-", "dontlog")]
    public string Password { get; set; }
}
// => ?dontlog-Password=...

Serializing a value object via ToString():

By default a complex query parameter is flattened into its public properties. To instead send a single value using the object's ToString() under the parameter's own name, mark it with an explicit empty format [Query(Format = "")] (or [Query(TreatAsString = true)]):

[Get("/info")]
Task<string> GetInfo([Query(Format = "")] Size size); // => ?size=medium  (uses size.ToString())

Custom query keys with IQueryConverter<T>:

When a parameter shape cannot be flattened from its declared type (an object, a polymorphic base type, a Dictionary<string, object>), or you simply need full control over the emitted keys, implement IQueryConverter<T> and attach it to the parameter with [QueryConverter(typeof(...))]. The converter writes query pairs straight into the pooled builder, so it can emit nested bracket keys such as order[createdAt]=desc without [AliasAs("order[")]-style hacks. This is a source-generator-only feature (the reflection request builder walks the value's runtime type instead); implementations must be stateless and have a public parameterless constructor.

public sealed class SortOrderQueryConverter : IQueryConverter<IDictionary<string, string>>
{
    public void Flatten(
        IDictionary<string, string> value,
        string keyPrefix,          // the resolved [Query(Prefix)] for the parameter, or an empty string
        ref GeneratedQueryStringBuilder builder,
        RefitSettings settings)
    {
        foreach (var entry in value)
        {
            // AddPreEscapedKey appends the key verbatim, so the brackets stay literal while the value is
            // still escaped. Use builder.Add(key, value, false) to percent-encode the key as well.
            builder.AddPreEscapedKey($"{keyPrefix}order[{entry.Key}]", entry.Value, false);
        }
    }
}

[Get("/items")]
Task<List<Item>> GetItems(
    [QueryConverter(typeof(SortOrderQueryConverter))] IDictionary<string, string> order);

GetItems(new Dictionary<string, string> { ["createdAt"] = "desc", ["priority"] = "asc" });
>>> "/items?order[createdAt]=desc&order[priority]=asc"

Formatting URL Parameter Values with the UrlParameterFormatter

In Refit, the UrlParameterFormatter property within RefitSettings allows you to customize how parameter values are formatted in the URL. This can be particularly useful when you need to format dates, numbers, or other types in a specific manner that aligns with your API's expectations.

Using UrlParameterFormatter:

Assign a custom formatter that implements the IUrlParameterFormatter interface to the UrlParameterFormatter property.

public class CustomDateUrlParameterFormatter : IUrlParameterFormatter
{
    public string? Format(object? value, ICustomAttributeProvider attributeProvider, Type type)
    {
        if (value is DateTime dt)
        {
            return dt.ToString("yyyyMMdd");
        }

        return value?.ToString();
    }
}

var settings = new RefitSettings
{
    UrlParameterFormatter = new CustomDateUrlParameterFormatter()
};

In this example, a custom formatter is created for date values. Whenever a DateTime parameter is encountered, it formats the date as yyyyMMdd.

Formatting Dictionary Keys:

When dealing with dictionaries, it's important to note that keys are treated as values. If you need custom formatting for dictionary keys, you should use the UrlParameterFormatter as well.

For instance, if you have a dictionary parameter and you want to format its keys in a specific way, you can handle that in the custom formatter:

public class CustomDictionaryKeyFormatter : IUrlParameterFormatter
{
    public string? Format(object? value, ICustomAttributeProvider attributeProvider, Type type)
    {
        // Handle dictionary keys
        if (attributeProvider is PropertyInfo prop && prop.PropertyType.IsGenericType && prop.PropertyType.GetGenericTypeDefinition() == typeof(Dictionary<,>))
        {
            // Custom formatting logic for dictionary keys
            return value?.ToString().ToUpperInvariant();
        }

        return value?.ToString();
    }
}

var settings = new RefitSettings
{
    UrlParameterFormatter = new CustomDictionaryKeyFormatter()
};

In the above example, the dictionary keys will be converted to uppercase.

Body content

One of the parameters in your method can be used as the body, by using the Body attribute:

[Post("/users/new")]
Task CreateUser([Body] User user);

There are four possibilities for supplying the body data, depending on the type of the parameter:

  • If the type is Stream, the content will be streamed via StreamContent
  • If the type is string, the string will be used directly as the content unless [Body(BodySerializationMethod.Json)] is set which will send it as a StringContent
  • If the parameter has the attribute [Body(BodySerializationMethod.UrlEncoded)], the content will be URL-encoded (see form posts below)
  • If the parameter has the attribute [Body(BodySerializationMethod.JsonLines)], an enumerable body is sent as JSON Lines (see JSON Lines content below)
  • For all other types, the object will be serialized using the content serializer specified in RefitSettings (JSON is the default).

Buffering and the Content-Length header

By default, Refit streams the body content without buffering it. This means you can stream a file from disk, for example, without incurring the overhead of loading the whole file into memory. The downside of this is that no Content-Length header is set on the request. If your API needs you to send a Content-Length header with the request, you can disable this streaming behavior by setting the buffered argument of the [Body] attribute to true:

Task CreateUser([Body(buffered: true)] User user);

JSON content

JSON requests and responses are serialized/deserialized using an instance of the IHttpContentSerializer interface. Refit provides two implementations out of the box: SystemTextJsonContentSerializer (which is the default JSON serializer) and NewtonsoftJsonContentSerializer. The first uses System.Text.Json APIs and is focused on high performance and low memory usage, while the latter uses the known Newtonsoft.Json library and is more versatile and customizable. You can read more about the two serializers and the main differences between the two at this link.

The default System.Text.Json serializer uses camelCase property names, case-insensitive matching, and reads numbers from JSON strings (NumberHandling = AllowReadingFromString). Override any of these by starting from Refit's defaults, tweaking the JsonSerializerOptions, and passing them in:

var options = SystemTextJsonContentSerializer.GetDefaultJsonSerializerOptions();
options.NumberHandling = JsonNumberHandling.Strict; // opt out of reading numbers from strings

var settings = new RefitSettings(new SystemTextJsonContentSerializer(options));
Fast-path source-generated serialization

The default options add custom converters and set NumberHandling, which the System.Text.Json source generator does not support on its serialization fast-path, so the default serializer always uses the (slower) metadata logic. If you want the fast-path, start from GetFastPathJsonSerializerOptions() instead — it omits the converters and NumberHandling so the options stay fast-path eligible — and assign a source-generated TypeInfoResolver:

var options = SystemTextJsonContentSerializer.GetFastPathJsonSerializerOptions();
options.TypeInfoResolver = MyJsonContext.Default; // a JsonSerializerContext you declare

var settings = new RefitSettings(new SystemTextJsonContentSerializer(options));
[JsonSourceGenerationOptions(PropertyNamingPolicy = JsonKnownNamingPolicy.CamelCase)]
[JsonSerializable(typeof(MyRequest))]
internal partial class MyJsonContext : JsonSerializerContext;

Two caveats: dropping the converters means object-typed members are no longer inferred and enums are no longer written as camelCase strings (add your own converters only if you accept losing the fast-path); and the fast-path runs through the synchronous serialization primitives (SerializeToUtf8Bytes, Serialize(Utf8JsonWriter, ...)) — the one API that bypasses it is the built-in JsonSerializer.SerializeAsync(Stream, ...).

By default Refit serializes request bodies with JsonContent, which uses exactly that SerializeAsync(Stream) path, so the fast-path is not used even with fast-path-eligible options. To opt in, set RequestBodySerialization on RefitSettings to one of the synchronous modes, which serialize through the fast-path:

  • RequestBodySerializationMode.Buffered — serialize synchronously to UTF-8 bytes up front and send as a ByteArrayContent. Sets a Content-Length header; best for small-to-medium bodies.
  • RequestBodySerializationMode.Streamed — serialize through a Utf8JsonWriter written to the request stream and flushed asynchronously. No Content-Length, but peak memory is bounded (pooled writer chunks rather than the whole body); best for large uploads.
var options = SystemTextJsonContentSerializer.GetFastPathJsonSerializerOptions();
options.TypeInfoResolver = MyJsonContext.Default;

var settings = new RefitSettings(new SystemTextJsonContentSerializer(options))
{
    RequestBodySerialization = RequestBodySerializationMode.Buffered, // or .Streamed for large uploads
};

Both modes require the content serializer to implement ISynchronousContentSerializer (the default SystemTextJsonContentSerializer does); otherwise the body falls back to the default asynchronous serialization.

For instance, here is how to create a new RefitSettings instance using the Newtonsoft.Json-based serializer (you'll also need to add a PackageReference to Refit.Newtonsoft.Json):

var settings = new RefitSettings(new NewtonsoftJsonContentSerializer());

If you're using Newtonsoft.Json APIs, you can customize their behavior by setting the Newtonsoft.Json.JsonConvert.DefaultSettings property:

JsonConvert.DefaultSettings =
    () => new JsonSerializerSettings() {
        ContractResolver = new CamelCasePropertyNamesContractResolver(),
        Converters = {new StringEnumConverter()}
    };

// Serialized as: {"day":"Saturday"}
await PostSomeStuff(new { Day = DayOfWeek.Saturday });

As these are global settings they will affect your entire application. It might be beneficial to isolate the settings for calls to a particular API. When creating a Refit generated live interface, you may optionally pass a RefitSettings that will allow you to specify what serializer settings you would like. This allows you to have different serializer settings for separate APIs:

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com",
    new RefitSettings {
        ContentSerializer = new NewtonsoftJsonContentSerializer(
            new JsonSerializerSettings {
                ContractResolver = new SnakeCasePropertyNamesContractResolver()
        }
    )});

var otherApi = RestService.For<IOtherApi>("https://api.example.com",
    new RefitSettings {
        ContentSerializer = new NewtonsoftJsonContentSerializer(
            new JsonSerializerSettings {
                ContractResolver = new CamelCasePropertyNamesContractResolver()
        }
    )});

Property serialization/deserialization can be customized using Json.NET's JsonProperty attribute:

public class Foo
{
    // Works like [AliasAs("b")] would in form posts (see below)
    [JsonProperty(PropertyName="b")]
    public string Bar { get; set; }
}
JSON source generator

To apply the benefits of the new JSON source generator for System.Text.Json added in .NET 6, you can use SystemTextJsonContentSerializer with a custom instance of RefitSettings and JsonSerializerOptions:

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com",
    new RefitSettings {
        ContentSerializer = new SystemTextJsonContentSerializer(MyJsonSerializerContext.Default.Options)
    });

When using System.Text.Json polymorphism features such as [JsonDerivedType] / [JsonPolymorphic], Refit serializes request bodies using the declared Refit method parameter type rather than the boxed runtime object. This ensures type discriminators configured on the base contract are preserved in outgoing request payloads.

JSON Lines content

Some APIs accept a batch of records as JSON Lines (newline-delimited JSON), where each line is a self-contained JSON document. Mark an enumerable body parameter with [Body(BodySerializationMethod.JsonLines)] and Refit serializes each element with the configured content serializer, writing one document per line:

public interface IDocumentApi
{
    [Post("/collections/companies/documents/import")]
    Task ImportAsync([Body(BodySerializationMethod.JsonLines)] IEnumerable<Company> documents);
}

The request body is streamed (one serialized element per line, separated by \n) with a content type of application/x-ndjson. A non-enumerable value is sent as a single line. If you need a different content type (for example text/plain), set it with a static header or a dynamic header.

Streaming responses with IAsyncEnumerable<T>

Declare a method that returns IAsyncEnumerable<T> to consume a large response without buffering the whole body. Refit reads the response with HttpCompletionOption.ResponseHeadersRead and yields each element as it is deserialized:

public interface IDocumentApi
{
    [Get("/documents")]
    IAsyncEnumerable<Company> GetDocuments();
}

await foreach (var company in api.GetDocuments())
{
    // each item is yielded as soon as it is read off the wire
}

The frame format is auto-detected from the response content type: a content type of application/jsonl, application/x-ndjson, or application/x-jsonlines is read as JSON Lines (one value per line); anything else is read as a single streamed JSON array. To observe cancellation, add a CancellationToken parameter, or use WithCancellation(token) on the enumerable.

Streaming requires the configured ContentSerializer to implement IStreamingContentSerializer. The default SystemTextJsonContentSerializer does; a serializer that does not will throw NotSupportedException when the sequence is enumerated.

XML Content

XML requests and responses are serialized/deserialized using System.Xml.Serialization.XmlSerializer. By default, Refit will use JSON content serialization, to use XML content configure the ContentSerializer to use the XmlContentSerializer:

var gitHubApi = RestService.For<IXmlApi>("https://www.w3.org/XML",
    new RefitSettings {
        ContentSerializer = new XmlContentSerializer()
    });

Property serialization/deserialization can be customized using attributes found in the System.Xml.Serialization namespace:

    public class Foo
    {
        [XmlElement(Namespace = "https://www.w3.org/XML")]
        public string Bar { get; set; }
    }

The System.Xml.Serialization.XmlSerializer provides many options for serializing, those options can be set by providing an XmlContentSerializerSettings to the XmlContentSerializer constructor:

var gitHubApi = RestService.For<IXmlApi>("https://www.w3.org/XML",
    new RefitSettings {
        ContentSerializer = new XmlContentSerializer(
            new XmlContentSerializerSettings
            {
                XmlReaderWriterSettings = new XmlReaderWriterSettings()
                {
                    ReaderSettings = new XmlReaderSettings
                    {
                        IgnoreWhitespace = true
                    }
                }
            }
        )
    });

Form posts

For APIs that take form posts (i.e. serialized as application/x-www-form-urlencoded), initialize the Body attribute with BodySerializationMethod.UrlEncoded.

The parameter can be an IDictionary:

public interface IMeasurementProtocolApi
{
    [Post("/collect")]
    Task Collect([Body(BodySerializationMethod.UrlEncoded)] Dictionary<string, object> data);
}

var data = new Dictionary<string, object> {
    {"v", 1},
    {"tid", "UA-1234-5"},
    {"cid", new Guid("d1e9ea6b-2e8b-4699-93e0-0bcbd26c206c")},
    {"t", "event"},
};

// Serialized as: v=1&tid=UA-1234-5&cid=d1e9ea6b-2e8b-4699-93e0-0bcbd26c206c&t=event
await api.Collect(data);

Or you can just pass any object and all public, readable properties will be serialized as form fields in the request. This approach allows you to alias property names using [AliasAs("whatever")] which can help if the API has cryptic field names:

public interface IMeasurementProtocolApi
{
    [Post("/collect")]
    Task Collect([Body(BodySerializationMethod.UrlEncoded)] Measurement measurement);
}

public class Measurement
{
    // Properties can be read-only and [AliasAs] isn't required
    public int v { get { return 1; } }

    [AliasAs("tid")]
    public string WebPropertyId { get; set; }

    [AliasAs("cid")]
    public Guid ClientId { get; set; }

    [AliasAs("t")]
    public string Type { get; set; }

    public object IgnoreMe { private get; set; }
}

var measurement = new Measurement {
    WebPropertyId = "UA-1234-5",
    ClientId = new Guid("d1e9ea6b-2e8b-4699-93e0-0bcbd26c206c"),
    Type = "event"
};

// Serialized as: v=1&tid=UA-1234-5&cid=d1e9ea6b-2e8b-4699-93e0-0bcbd26c206c&t=event
await api.Collect(measurement);

If you have a type that has [JsonProperty(PropertyName)] attributes setting property aliases, Refit will use those too ([AliasAs] will take precedence where you have both). This means that the following type will serialize as one=value1&two=value2:

public class SomeObject
{
    [JsonProperty(PropertyName = "one")]
    public string FirstProperty { get; set; }

    [JsonProperty(PropertyName = "notTwo")]
    [AliasAs("two")]
    public string SecondProperty { get; set; }
}

NOTE: This use of AliasAs applies to querystring parameters and form body posts, but not to response objects; for aliasing fields on response objects, you'll still need to use [JsonProperty("full-property-name")].

Sending null values

By default, a property whose value is null is omitted from the form body (and from object querystrings). To send an explicit empty value (key=) for a null property instead of omitting it, set SerializeNull on its [Query] attribute:

public class Measurement
{
    [Query(SerializeNull = true)]
    public string? Note { get; set; }
}

// With Note = null, serialized as: ...&Note=
Reflection-free form serialization

With generated request building on (the default), Refit flattens strongly-typed form bodies at compile time, so no reflection runs at request time. This covers a concrete class or struct serialized with the built-in SystemTextJsonContentSerializer. An object, an IDictionary, a collection, or a custom IHttpContentSerializer falls back to the reflection path, which produces identical output.

A body whose properties are all simple scalars (strings, numbers, enums, other IFormattable values) takes a faster straight-line path: the generator unrolls each field directly, with no descriptor array, getter delegates, or boxing. A body with a collection or complex property keeps the descriptor path. All three paths produce identical output.

Setting request headers

Static headers

You can set one or more static request headers for a request applying a Headers attribute to the method:

[Headers("User-Agent: Awesome Octocat App")]
[Get("/users/{user}")]
Task<User> GetUser(string user);

Static headers can also be added to every request in the API by applying the Headers attribute to the interface:

[Headers("User-Agent: Awesome Octocat App")]
public interface IGitHubApi
{
    [Get("/users/{user}")]
    Task<User> GetUser(string user);

    [Post("/users/new")]
    Task CreateUser([Body] User user);
}

Dynamic headers

If the content of the header needs to be set at runtime, you can add a header with a dynamic value to a request by applying a Header attribute to a parameter:

[Get("/users/{user}")]
Task<User> GetUser(string user, [Header("Authorization")] string authorization);

// Will add the header "Authorization: token OAUTH-TOKEN" to the request
var user = await GetUser("octocat", "token OAUTH-TOKEN");

Adding an Authorization header is such a common use case that you can add an access token to a request by applying an Authorize attribute to a parameter and optionally specifying the scheme:

[Get("/users/{user}")]
Task<User> GetUser(string user, [Authorize("Bearer")] string token);

// Will add the header "Authorization: Bearer OAUTH-TOKEN}" to the request
var user = await GetUser("octocat", "OAUTH-TOKEN");

//note: the scheme defaults to Bearer if none provided

If you need to set multiple headers at runtime, you can add a IDictionary<string, string> and apply a HeaderCollection attribute to the parameter and it will inject the headers into the request:

[Get("/users/{user}")]
Task<User> GetUser(string user, [HeaderCollection] IDictionary<string, string> headers);

var headers = new Dictionary<string, string> {{"Authorization","Bearer tokenGoesHere"}, {"X-Tenant-Id","123"}};
var user = await GetUser("octocat", headers);

Bearer Authentication

Most APIs need some sort of Authentication. The most common is OAuth Bearer authentication. A header is added to each request of the form: Authorization: Bearer <token>. Refit makes it easy to insert your logic to get the token however your app needs, so you don't have to pass a token into each method.

  1. Add [Headers("Authorization: Bearer")] to the interface or methods which need the token.
  2. Set AuthorizationHeaderValueGetter in the RefitSettings instance. Refit will call your delegate each time it needs to obtain the token, so it's a good idea for your mechanism to cache the token value for some period within the token lifetime.

AuthorizationHeaderValueGetter works whether you create clients with RestService.For<T>("https://...") or supply your own HttpClient via RestService.For<T>(httpClient, settings). If your API methods accept a CancellationToken, that token is propagated to the getter delegate.

If your getter returns null, an empty string, or whitespace, Refit omits the Authorization header for that request instead of sending a blank Authorization: <scheme> value. This lets a single client make both authenticated and anonymous calls: return a token when you have one, and return an empty string to skip auth for that request. Omitting the Authorization placeholder entirely (no [Headers("Authorization: Bearer")]) skips auth for the whole method.

Scoped (per-request) authorization tokens with dependency injection

When you register a client through Refit.HttpClientFactory, you can resolve the token from dependency injection with AddAuthorizationHeaderValueProvider:

services.AddRefitClient<IMyApi>()
    .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"))
    .AddAuthorizationHeaderValueProvider((serviceProvider, request, cancellationToken) =>
    {
        var tokenService = serviceProvider.GetRequiredService<IMyTokenService>();
        return new ValueTask<string>(tokenService.GetTokenForCurrentRequest());
    });

The delegate receives an IServiceProvider, the outgoing HttpRequestMessage, and a CancellationToken, and returns the token to place in the Authorization header (returning null/empty/whitespace skips auth for that request, exactly like AuthorizationHeaderValueGetter).

Caveat: IHttpClientFactory pools message handlers for their configured lifetime, so a scoped service captured directly by a handler would bleed across requests. To keep the provider correctly per-request, this extension creates a fresh DI scope for every request and resolves your delegate from that scope's IServiceProvider, disposing the scope when the request completes. True per-request isolation therefore relies on either per-request state you read from the request argument, or ambient AsyncLocal-based state (such as a host-registered IHttpContextAccessor) that flows into the fresh scope. No Microsoft.AspNetCore.* reference is required.

Reducing header boilerplate with DelegatingHandlers (Authorization headers worked example)

Although we make provisions for adding dynamic headers at runtime directly in Refit, most use-cases would likely benefit from registering a custom DelegatingHandler in order to inject the headers as part of the HttpClient middleware pipeline thus removing the need to add lots of [Header] or [HeaderCollection] attributes.

In the example above we are leveraging a [HeaderCollection] parameter to inject an Authorization and X-Tenant-Id header. This is quite a common scenario if you are integrating with a 3rd party that uses OAuth2. While it's ok for the occasional endpoint, it would be quite cumbersome if we had to add that boilerplate to every method in our interface.

In this example we will assume our application is a multi-tenant application that is able to pull information about a tenant through some interface ITenantProvider and has a data store IAuthTokenStore that can be used to retrieve an auth token to attach to the outbound request.

 //Custom delegating handler for adding Auth headers to outbound requests
 class AuthHeaderHandler : DelegatingHandler
 {
     private readonly ITenantProvider tenantProvider;
     private readonly IAuthTokenStore authTokenStore;

    public AuthHeaderHandler(ITenantProvider tenantProvider, IAuthTokenStore authTokenStore)
    {
         this.tenantProvider = tenantProvider ?? throw new ArgumentNullException(nameof(tenantProvider));
         this.authTokenStore = authTokenStore ?? throw new ArgumentNullException(nameof(authTokenStore));
         // InnerHandler must be left as null when using DI, but must be assigned a value when
         // using RestService.For<IMyApi>
         // InnerHandler = new HttpClientHandler();
    }

    protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
    {
        var token = await authTokenStore.GetToken();

        //potentially refresh token here if it has expired etc.

        request.Headers.Authorization = new AuthenticationHeaderValue("Bearer", token);
        request.Headers.Add("X-Tenant-Id", tenantProvider.GetTenantId());

        return await base.SendAsync(request, cancellationToken).ConfigureAwait(false);
    }
}

//Startup.cs
public void ConfigureServices(IServiceCollection services)
{
    services.AddTransient<ITenantProvider, TenantProvider>();
    services.AddTransient<IAuthTokenStore, AuthTokenStore>();
    services.AddTransient<AuthHeaderHandler>();

    //this will add our refit api implementation with an HttpClient
    //that is configured to add auth headers to all requests

    //note: AddRefitClient<T> requires a reference to Refit.HttpClientFactory
    //note: the order of delegating handlers is important and they run in the order they are added!

    services.AddRefitClient<ISomeThirdPartyApi>()
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"))
        .AddHttpMessageHandler<AuthHeaderHandler>();
        //you could add Polly here to handle HTTP 429 / HTTP 503 etc
}

//Your application code
public class SomeImportantBusinessLogic
{
    private ISomeThirdPartyApi thirdPartyApi;

    public SomeImportantBusinessLogic(ISomeThirdPartyApi thirdPartyApi)
    {
        this.thirdPartyApi = thirdPartyApi;
    }

    public async Task DoStuffWithUser(string username)
    {
        var user = await thirdPartyApi.GetUser(username);
        //do your thing
    }
}

If you aren't using dependency injection then you could achieve the same thing by doing something like this:

var api = RestService.For<ISomeThirdPartyApi>(new HttpClient(new AuthHeaderHandler(tenantProvider, authTokenStore))
    {
        BaseAddress = new Uri("https://api.example.com")
    }
);

var user = await thirdPartyApi.GetUser(username);
//do your thing

Redefining headers

Unlike Retrofit, where headers do not overwrite each other and are all added to the request regardless of how many times the same header is defined, Refit takes a similar approach to the approach ASP.NET MVC takes with action filters — redefining a header will replace it, in the following order of precedence:

  • Headers attribute on the interface (lowest priority)
  • Headers attribute on the method
  • Header attribute or HeaderCollection attribute on a method parameter (highest priority)
[Headers("X-Emoji: :rocket:")]
public interface IGitHubApi
{
    [Get("/users/list")]
    Task<List> GetUsers();

    [Get("/users/{user}")]
    [Headers("X-Emoji: :smile_cat:")]
    Task<User> GetUser(string user);

    [Post("/users/new")]
    [Headers("X-Emoji: :metal:")]
    Task CreateUser([Body] User user, [Header("X-Emoji")] string emoji);
}

// X-Emoji: :rocket:
var users = await GetUsers();

// X-Emoji: :smile_cat:
var user = await GetUser("octocat");

// X-Emoji: :trollface:
await CreateUser(user, ":trollface:");

Note: This redefining behavior only applies to headers with the same name. Headers with different names are not replaced. The following code will result in all headers being included:

[Headers("Header-A: 1")]
public interface ISomeApi
{
    [Headers("Header-B: 2")]
    [Post("/post")]
    Task PostTheThing([Header("Header-C")] int c);
}

// Header-A: 1
// Header-B: 2
// Header-C: 3
var user = await api.PostTheThing(3);

Removing headers

Headers defined on an interface or method can be removed by redefining a static header without a value (i.e. without : <value>) or passing null for a dynamic header. Empty strings will be included as empty headers.

[Headers("X-Emoji: :rocket:")]
public interface IGitHubApi
{
    [Get("/users/list")]
    [Headers("X-Emoji")] // Remove the X-Emoji header
    Task<List> GetUsers();

    [Get("/users/{user}")]
    [Headers("X-Emoji:")] // Redefine the X-Emoji header as empty
    Task<User> GetUser(string user);

    [Post("/users/new")]
    Task CreateUser([Body] User user, [Header("X-Emoji")] string emoji);
}

// No X-Emoji header
var users = await GetUsers();

// X-Emoji:
var user = await GetUser("octocat");

// No X-Emoji header
await CreateUser(user, null);

// X-Emoji:
await CreateUser(user, "");

Passing state into DelegatingHandlers

If there is runtime state that you need to pass to a DelegatingHandler you can add a property with a dynamic value to the underlying HttpRequestMessage.Properties by applying a Property attribute to a parameter:

public interface IGitHubApi
{
    [Post("/users/new")]
    Task CreateUser([Body] User user, [Property("SomeKey")] string someValue);

    [Post("/users/new")]
    Task CreateUser([Body] User user, [Property] string someOtherKey);
}

The attribute constructor optionally takes a string which becomes the key in the HttpRequestMessage.Properties dictionary. If no key is explicitly defined then the name of the parameter becomes the key. If a key is defined multiple times the value in HttpRequestMessage.Properties will be overwritten. The parameter itself can be any object. Properties can be accessed inside a DelegatingHandler as follows:

⚠️ Important for IHttpClientFactory users: DelegatingHandler instances are pooled and can live longer than a single request scope. Avoid reading per-request state from services that may be scoped/cached across handler lifetimes ( for example a tenant/customer resolver stored on the handler). For per-request values like CustomerId, pass the value through [Property] so each request carries its own state.

class RequestPropertyHandler : DelegatingHandler
{
    public RequestPropertyHandler(HttpMessageHandler innerHandler = null) : base(innerHandler ?? new HttpClientHandler()) {}

    protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
    {
        // See if the request has a the property
        if(request.Properties.ContainsKey("SomeKey"))
        {
            var someProperty = request.Properties["SomeKey"];
            //do stuff
        }

        if(request.Properties.ContainsKey("someOtherKey"))
        {
            var someOtherProperty = request.Properties["someOtherKey"];
            //do stuff
        }

        return await base.SendAsync(request, cancellationToken).ConfigureAwait(false);
    }
}

Note: in .NET 5 HttpRequestMessage.Properties has been marked Obsolete and Refit will instead populate the value into the new HttpRequestMessage.Options.

Support for Polly and Polly.Context

Because Refit supports HttpClientFactory it is possible to configure Polly policies on your HttpClient. If your policy makes use of Polly.Context this can be passed via Refit by adding [Property("PolicyExecutionContext")] Polly.Context context as behind the scenes Polly.Context is simply stored in HttpRequestMessage.Properties under the key PolicyExecutionContext and is of type Polly.Context. It's only recommended to pass the Polly.Context this way if your use case requires that the Polly.Context be initialized with dynamic content only known at runtime. If your Polly.Context only requires the same content every time (e.g an ILogger that you want to use to log from inside your policies) a cleaner approach is to inject the Polly.Context via a DelegatingHandler as described in #801

Target Interface Type and method info

There may be times when you want to know what the target interface type is of the Refit instance. An example is where you have a derived interface that implements a common base like this:

public interface IGetAPI<TEntity>
{
    [Get("/{key}")]
    Task<TEntity> Get(long key);
}

public interface IUsersAPI : IGetAPI<User>
{
}

public interface IOrdersAPI : IGetAPI<Order>
{
}

You can access the concrete type of the interface for use in a handler, such as to alter the URL of the request:

class RequestPropertyHandler : DelegatingHandler
{
    public RequestPropertyHandler(HttpMessageHandler innerHandler = null) : base(innerHandler ?? new HttpClientHandler()) {}

    protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
    {
        // Get the type of the target interface
        Type interfaceType = (Type)request.Properties[HttpMessageRequestOptions.InterfaceType];

        var builder = new UriBuilder(request.RequestUri);
        // Alter the Path in some way based on the interface or an attribute on it
        builder.Path = $"/{interfaceType.Name}{builder.Path}";
        // Set the new Uri on the outgoing message
        request.RequestUri = builder.Uri;

        return await base.SendAsync(request, cancellationToken).ConfigureAwait(false);
    }
}

The full method information (RestMethodInfo) is also always available in the request options. The RestMethodInfo contains more information about the method being called such as the full MethodInfo when using reflection is needed:

class RequestPropertyHandler : DelegatingHandler
{
    public RequestPropertyHandler(HttpMessageHandler innerHandler = null) : base(innerHandler ?? new HttpClientHandler()) {}

    protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
    {
        // Get the method info
        if (request.Options.TryGetValue(new HttpRequestOptionsKey<RestMethodInfo>(HttpRequestMessageOptions.RestMethodInfo), out RestMethodInfo restMethodInfo))
        {
            var builder = new UriBuilder(request.RequestUri);
            // Alter the Path in some way based on the method info or an attribute on it
            builder.Path = $"/{restMethodInfo.MethodInfo.Name}{builder.Path}";
            // Set the new Uri on the outgoing message
            request.RequestUri = builder.Uri;
        }

        return await base.SendAsync(request, cancellationToken).ConfigureAwait(false);
    }
}

Note: in .NET 5 HttpRequestMessage.Properties has been marked Obsolete and Refit will instead populate the value into the new HttpRequestMessage.Options. Refit provides HttpRequestMessageOptions.InterfaceType and HttpRequestMessageOptions.RestMethodInfo to respectively access the interface type and REST method info from the options.

Inspecting the current call's arguments

Set RefitSettings.CaptureMethodArguments = true to expose the current call's argument values to a DelegatingHandler via HttpRequestMessageOptions.MethodArguments. The stored value is an object?[] holding the boxed arguments in the method's declared parameter order, including any CancellationToken, so it lines up 1:1 with the reflected parameter list. This mirrors Retrofit's Invocation.arguments.

var settings = new RefitSettings { CaptureMethodArguments = true };

class ArgumentLoggingHandler : DelegatingHandler
{
    public ArgumentLoggingHandler(HttpMessageHandler innerHandler = null) : base(innerHandler ?? new HttpClientHandler()) {}

    protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request, CancellationToken cancellationToken)
    {
        if (request.Options.TryGetValue(new HttpRequestOptionsKey<object?[]>(HttpRequestMessageOptions.MethodArguments), out var arguments))
        {
            // Reflection path: pair the values with RestMethodInfo.MethodInfo.GetParameters() to recover parameter names.
            if (request.Options.TryGetValue(new HttpRequestOptionsKey<RestMethodInfo>(HttpRequestMessageOptions.RestMethodInfo), out var restMethodInfo))
            {
                var parameters = restMethodInfo.MethodInfo.GetParameters();
                for (var i = 0; i < arguments.Length; i++)
                {
                    Console.WriteLine($"{parameters[i].Name} = {arguments[i]}");
                }
            }
        }

        return await base.SendAsync(request, cancellationToken).ConfigureAwait(false);
    }
}

It is off by default: enabling it boxes and retains the arguments on every request (and therefore on any resulting ApiException) for the lifetime of that request, so it adds a per-call allocation, keeps otherwise-collectable arguments alive, and the captured values frequently contain credentials or PII — the same trade-offs as CaptureRequestContent. Use RefitSettings.ExceptionRedactor to scrub the values before an exception reaches a logging or telemetry pipeline.

The captured values are positional. Under the source-generated request path they are supplied as a bare object?[] with no parameter names attached, so match them against your interface method's declared parameter order. RestMethodInfo (with its MethodInfo.GetParameters()) is only published on the reflection path today, so the name-recovery shown above applies there; the generated path does not currently surface RestMethodInfo.

Multipart uploads

Methods decorated with Multipart attribute will be submitted with multipart content type. At this time, multipart methods support the following parameter types:

  • string (parameter name will be used as name and string value as value)
  • Guid, DateTime, DateTimeOffset, TimeSpan (and DateOnly / TimeOnly on supported targets) — sent as their plain text form (not JSON-quoted)
  • byte array
  • Stream
  • FileInfo

Name of the field in the multipart data priority precedence:

  • multipartItem.Name if specified and not null (optional); dynamic, allows naming form data part at execution time.
  • [AliasAs] attribute (optional) that decorate the streamPart parameter in the method signature (see below); static, defined in code.
  • MultipartItem parameter name (default) as defined in the method signature; static, defined in code.

A custom boundary can be specified with an optional string parameter to the Multipart attribute. If left empty, this defaults to ----MyGreatBoundary.

To specify the file name and content type for byte array (byte[]), Stream and FileInfo parameters, use of a wrapper class is required. The wrapper classes for these types are ByteArrayPart, StreamPart and FileInfoPart.

public interface ISomeApi
{
    [Multipart]
    [Post("/users/{id}/photo")]
    Task UploadPhoto(int id, [AliasAs("myPhoto")] StreamPart stream);
}

To pass a Stream to this method, construct a StreamPart object like so:

someApiInstance.UploadPhoto(id, new StreamPart(myPhotoStream, "photo.jpg", "image/jpeg"));

Note: The AttachmentName attribute that was previously described in this section has been deprecated and its use is not recommended.

Retrieving the response

Note that in Refit unlike in Retrofit, there is no option for a synchronous network request - all requests must be async, either via Task or via IObservable. There is also no option to create an async method via a Callback parameter unlike Retrofit, because we live in the async/await future.

Similarly to how body content changes via the parameter type, the return type will determine the content returned.

Returning Task without a type parameter will discard the content and solely tell you whether or not the call succeeded:

[Post("/users/new")]
Task CreateUser([Body] User user);

// This will throw if the network call fails
await CreateUser(someUser);

If the type parameter is 'HttpResponseMessage' or 'string', the raw response message or the content as a string will be returned respectively.

// Returns the content as a string (i.e. the JSON data)
[Get("/users/{user}")]
Task<string> GetUser(string user);

// Returns the raw response, as an IObservable that can be used with the
// Reactive Extensions
[Get("/users/{user}")]
IObservable<HttpResponseMessage> GetUser(string user);

There is also a generic wrapper class called ApiResponse<T> that can be used as a return type. Using this class as a return type allows you to retrieve not just the content as an object, but also any metadata associated with the request/response. This includes information such as response headers, the http status code and reason phrase (e.g. 404 Not Found), the response version, the original request message that was sent and in the case of an error, an ApiException object containing details of the error. Following are some examples of how you can retrieve the response metadata.

//Returns the content within a wrapper class containing metadata about the request/response
[Get("/users/{user}")]
Task<ApiResponse<User>> GetUser(string user);

//Calling the API
var response = await gitHubApi.GetUser("octocat");

//Determining if a success status code was received and there wasn't any other error
//(for example, during content deserialization)
if(response.IsSuccessful)
{
    //YAY! Do the thing...
}

if (response.IsReceived)
{
    //Getting the status code (returns a value from the System.Net.HttpStatusCode enumeration)
    var httpStatus = response.StatusCode;

    //Retrieving a well-known header value (e.g. "Server" header)
    var serverHeaderValue = response.Headers.Server != null ? response.Headers.Server.ToString() : string.Empty;

    //Retrieving a custom header value
    var customHeaderValue = string.Join(',', response.Headers.GetValues("A-Custom-Header"));

    //Looping through all the headers
    foreach(var header in response.Headers)
    {
        var headerName = header.Key;
        var headerValue = string.Join(',', header.Value);
    }

    //Finally, retrieving the content in the response body as a strongly-typed object
    var user = response.Content;
}

A successful response does not imply a response body. Per RFC 9110 a 204 (No Content), a 304 (Not Modified), a response to HEAD, or any 2xx with a zero-length or literal null body deserializes to Content == null without an error — so IsSuccessful / IsSuccessStatusCode being true tells you the request succeeded, not that Content is present. Use one of the content-presence members below when you need the deserialized body:

  • HasContent — the covariance-safe way to null-check Content on its own; when true the compiler knows Content is non-null.
  • IsSuccessfulWithContent — a single check that combines success and content presence (IsSuccessful && Content is not null); when true the compiler knows Content is non-null.
if (response.HasContent)
{
    // response.Content is non-null here
    Process(response.Content);
}

// Or, when you want "succeeded and has a body" in one narrowing check:
if (response.IsSuccessfulWithContent)
{
    // response.Content is non-null here
    Process(response.Content);
}

EnsureSuccessStatusCodeAsync() and EnsureSuccessfulAsync() are available directly on IApiResponse<T> (not just the concrete ApiResponse<T>), throwing the captured ApiException / ApiRequestException when the request was not successful:

IApiResponse<User> response = await gitHubApi.GetUser("octocat");
await response.EnsureSuccessStatusCodeAsync();

Do I need to dispose the response?

ApiResponse<T> (and IApiResponse / IApiResponse<T>) implement IDisposable, but for the common case you do not need to dispose them and you are not leaking sockets. For every return type except the streaming ones below, Refit buffers the body into memory, returns the connection to the pool, and disposes the underlying HttpResponseMessage before handing the result back to you. Calling Dispose() (or a using) on those responses is a harmless, idempotent no-op — fine to keep to satisfy analyzers, but not required.

Disposal genuinely matters only when you ask Refit for the live response, where the body is not buffered and you own the connection until you dispose it:

  • HttpResponseMessage, HttpContent, Stream
  • ApiResponse<HttpResponseMessage>, ApiResponse<HttpContent>, ApiResponse<Stream>
// Streaming: you own the response — dispose it (e.g. with `using`)
[Get("/files/{id}")]
Task<ApiResponse<Stream>> DownloadAsync(string id);

using var response = await api.DownloadAsync("42");
await using var stream = response.Content; // released when the response is disposed

The rule is decided in RestMethodInfoInternal.DetermineIfResponseMustBeDisposed: those three types (and their ApiResponse<> wrappers) are the only ones Refit does not auto-dispose for you.

Custom return types (IReturnTypeAdapter)

Beyond the built-in shapes, you can teach Refit to surface any return type by implementing IReturnTypeAdapter<TReturn, TResult>. TReturn is the type your interface method returns; TResult is what the HTTP call materializes (the deserialized body). The single Adapt method receives a deferred call and returns the wrapper:

// A cold, deferred call that runs only when invoked.
public sealed class DeferredCall<T>(Func<CancellationToken, Task<T>> invoke)
{
    public Task<T> InvokeAsync(CancellationToken ct = default) => invoke(ct);
}

public sealed class DeferredCallAdapter<T> : IReturnTypeAdapter<DeferredCall<T>, T>
{
    public DeferredCall<T> Adapt(Func<CancellationToken, Task<T>> invoke) => new(invoke);
}

public interface IUserApi
{
    [Get("/users/{id}")]
    DeferredCall<User> GetUser(int id); // surfaced by the adapter above
}
  • Generated (Native AOT / trimming friendly): the source generator discovers adapters declared in your project at compile time and emits a direct Adapt call — no reflection or registration needed.
  • Reflection (RestService.For without the generator, or adapters from another assembly): register the adapter type on the settings — settings.ReturnTypeAdapters.Add(typeof(DeferredCallAdapter<>)).

A generic adapter's single type parameter is treated as the wrapped result type (Adapter<T> : IReturnTypeAdapter<Wrapper<T>, T>), so Wrapper<User> closes it over User. Adapters must have a public parameterless constructor. The generated path builds the request eagerly and captures it, so a generated deferred call is single-use; the reflection path rebuilds the request on each invocation, so it can re-run.

Using generic interfaces

When using something like ASP.NET Web API, it's a fairly common pattern to have a whole stack of CRUD REST services. Refit now supports these, allowing you to define a single API interface with a generic type:

public interface IReallyExcitingCrudApi<T, in TKey> where T : class
{
    [Post("")]
    Task<T> Create([Body] T payload);

    [Get("")]
    Task<List<T>> ReadAll();

    [Get("/{key}")]
    Task<T> ReadOne(TKey key);

    [Put("/{key}")]
    Task Update(TKey key, [Body]T payload);

    [Delete("/{key}")]
    Task Delete(TKey key);
}

Which can be used like this:

// The "/users" part here is kind of important if you want it to work for more
// than one type (unless you have a different domain for each type)
var api = RestService.For<IReallyExcitingCrudApi<User, string>>("http://api.example.com/users");

Interface inheritance

When multiple services that need to be kept separate share a number of APIs, it is possible to leverage interface inheritance to avoid having to define the same Refit methods multiple times in different services:

public interface IBaseService
{
    [Get("/resources")]
    Task<Resource> GetResource(string id);
}

public interface IDerivedServiceA : IBaseService
{
    [Delete("/resources")]
    Task DeleteResource(string id);
}

public interface IDerivedServiceB : IBaseService
{
    [Post("/resources")]
    Task<string> AddResource([Body] Resource resource);
}

In this example, the IDerivedServiceA interface will expose both the GetResource and DeleteResource APIs, while IDerivedServiceB will expose GetResource and AddResource.

Composing multiple APIs into one client

The generator walks the full interface hierarchy, so you can split a large API by resource area into focused interfaces and then aggregate them into a single client that declares nothing of its own:

public interface IUsersApi
{
    [Get("/users/{user}")]
    Task<User> GetUser(string user);
}

public interface IReposApi
{
    [Get("/users/{user}/repos")]
    Task<List<Repo>> GetRepos(string user);
}

// The aggregate client has no members of its own; it just composes the two APIs.
public interface IGitHubApi : IUsersApi, IReposApi;

The composed client exposes every method from both base interfaces:

var api = RestService.For<IGitHubApi>("https://api.github.com");

var user = await api.GetUser("octocat");     // from IUsersApi
var repos = await api.GetRepos("octocat");   // from IReposApi

It works the same way with HttpClientFactory and dependency injection:

builder.Services
    .AddRefitClient<IGitHubApi>()
    .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.github.com"));

Each sub-interface (IUsersApi, IReposApi) can also be registered independently if some consumers only need one area, and the header-attribute precedence rules described just below still apply to the composed client.

Headers inheritance

When using inheritance, existing header attributes will be passed along as well, and the inner-most ones will have precedence:

[Headers("User-Agent: AAA")]
public interface IAmInterfaceA
{
    [Get("/get?result=Ping")]
    Task<string> Ping();
}

[Headers("User-Agent: BBB")]
public interface IAmInterfaceB : IAmInterfaceA
{
    [Get("/get?result=Pang")]
    [Headers("User-Agent: PANG")]
    Task<string> Pang();

    [Get("/get?result=Foo")]
    Task<string> Foo();
}

Here, IAmInterfaceB.Pang() will use PANG as its user agent, while IAmInterfaceB.Foo and IAmInterfaceB.Ping will use BBB. Note that if IAmInterfaceB didn't have a header attribute, Foo would then use the AAA value inherited from IAmInterfaceA. If an interface is inheriting more than one interface, the order of precedence is the same as the one in which the inherited interfaces are declared:

public interface IAmInterfaceC : IAmInterfaceA, IAmInterfaceB
{
    [Get("/get?result=Foo")]
    Task<string> Foo();
}

Here IAmInterfaceC.Foo would use the header attribute inherited from IAmInterfaceA, if present, or the one inherited from IAmInterfaceB, and so on for all the declared interfaces.

Default Interface Methods

Starting with C# 8.0, default interface methods (a.k.a. DIMs) can be defined on interfaces. Refit interfaces can provide additional logic using DIMs, optionally combined with private and/or static helper methods:

public interface IApiClient
{
    // implemented by Refit but not exposed publicly
    [Get("/get")]
    internal Task<string> GetInternal();
    // Publicly available with added logic applied to the result from the API call
    public async Task<string> Get()
        => FormatResponse(await GetInternal());
    private static String FormatResponse(string response)
        => $"The response is: {response}";
}

The type generated by Refit will implement the method IApiClient.GetInternal. If additional logic is required immediately before or after its invocation, it shouldn't be exposed directly and can thus be hidden from consumers by being marked as internal. The default interface method IApiClient.Get will be inherited by all types implementing IApiClient, including - of course - the type generated by Refit. Consumers of the IApiClient will call the public Get method and profit from the additional logic provided in its implementation (optionally, in this case, with the help of the private static helper FormatResponse). To support runtimes without DIM-support (.NET Core 2.x and below or .NET Standard 2.0 and below), two additional types would be required for the same solution.

internal interface IApiClientInternal
{
    [Get("/get")]
    Task<string> Get();
}
public interface IApiClient
{
    public Task<string> Get();
}
internal class ApiClient : IApiClient
{
    private readonly IApiClientInternal client;
    public ApiClient(IApiClientInternal client) => this.client = client;
    public async Task<string> Get()
        => FormatResponse(await client.Get());
    private static String FormatResponse(string response)
        => $"The response is: {response}";
}

Using HttpClientFactory

Refit has first class support for IHttpClientFactory. Add a reference to Refit.HttpClientFactory and call the provided extension method in your ConfigureServices method to configure your Refit interface:

services.AddRefitClient<IWebApi>()
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));
        // Add additional IHttpClientBuilder chained methods as required here:
        // .AddHttpMessageHandler<MyHandler>()
        // .SetHandlerLifetime(TimeSpan.FromMinutes(2));

Optionally, a RefitSettings object can be included:

var settings = new RefitSettings();
// Configure refit settings here

services.AddRefitClient<IWebApi>(settings)
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));
        // Add additional IHttpClientBuilder chained methods as required here:
        // .AddHttpMessageHandler<MyHandler>()
        // .SetHandlerLifetime(TimeSpan.FromMinutes(2));

// or injected from the container
services.AddRefitClient<IWebApi>(provider => new RefitSettings() { /* configure settings */ })
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));
        // Add additional IHttpClientBuilder chained methods as required here:
        // .AddHttpMessageHandler<MyHandler>()
        // .SetHandlerLifetime(TimeSpan.FromMinutes(2));

Note that some of the properties of RefitSettings will be ignored because the HttpClient and HttpClientHandlers will be managed by the HttpClientFactory instead of Refit.

Refit registers each client with IHttpClientFactory under a deterministic name. You can compute that same name with UniqueName.ForType<T>(), for example to configure the underlying named HttpClient directly:

services.AddHttpClient(UniqueName.ForType<IWebApi>())
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));

If you would rather use a short, human-readable name, pass httpClientName to any AddRefitClient overload. That name becomes the IHttpClientFactory client name and the client's default ILogger logging category, so it shortens both without changing the assembly-qualified default that keeps registrations unique:

services.AddRefitClient<IWebApi>(settings: null, httpClientName: "web-api")
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));

You can then get the api interface using constructor injection:

public class HomeController : Controller
{
    public HomeController(IWebApi webApi)
    {
        _webApi = webApi;
    }

    private readonly IWebApi _webApi;

    public async Task<IActionResult> Index(CancellationToken cancellationToken)
    {
        var thing = await _webApi.GetSomethingWeNeed(cancellationToken);
        return View(thing);
    }
}

Sharing one connection pool across multiple interfaces

If you split one API across several interfaces (for example ISomeSiteAuth and ISomeSiteData) but want them to share a single underlying handler and connection pool, register them under the same client name. Interfaces that resolve the same named HttpClient share the same IHttpClientFactory-managed, lifetime-rotated handler:

services.AddRefitClient<ISomeSiteAuth>((RefitSettings?)null, "somesite")
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));

services.AddRefitClient<ISomeSiteData>((RefitSettings?)null, "somesite")
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));

You do not need to do this for correctness. By default each interface gets its own factory-managed named client, and IHttpClientFactory already pools and rotates the underlying handlers, so there is no socket exhaustion or DNS-staleness concern. Sharing a name simply collapses the interfaces onto one pool (and one shared ConfigureHttpClient/handler configuration) when they target the same host.

Outside DI you can share a single HttpClient directly by passing the same instance to multiple RestService.For<T>(client) calls (see Providing a custom HttpClient below). Prefer sharing the pooled handler over caching long-lived HttpClient instances yourself, so handler rotation and DNS refresh keep working.

Providing a custom HttpClient

You can supply a custom HttpClient instance by simply passing it as a parameter to the RestService.For<T> method:

RestService.For<ISomeApi>(new HttpClient()
{
    BaseAddress = new Uri("https://www.someapi.com/api/")
});

However, when supplying a custom HttpClient instance, HttpMessageHandlerFactory will not be used because you already control the handler pipeline.

AuthorizationHeaderValueGetter does still work with RestService.For<T>(httpClient, settings) when the request includes an Authorization header placeholder (for example [Headers("Authorization: Bearer")]).

If you still want to be able to configure the HtttpClient instance that Refit provides while still making use of the above settings, simply expose the HttpClient on the API interface:

interface ISomeApi
{
    // This will automagically be populated by Refit if the property exists
    HttpClient Client { get; }

    [Headers("Authorization: Bearer")]
    [Get("/endpoint")]
    Task<string> SomeApiEndpoint();
}

Then, after creating the REST service, you can set any HttpClient property you want, e.g. Timeout:

SomeApi = RestService.For<ISomeApi>("https://www.someapi.com/api/", new RefitSettings()
{
    AuthorizationHeaderValueGetter = (rq, ct) => GetTokenAsync()
});

SomeApi.Client.Timeout = timeout;

Native AoT / trimming guidance

Refit's recommended source-generator-first setup for Native AoT and trimmed applications is:

  1. Use normal Refit interfaces so the Refit source generator produces the client implementation at build time.
  2. Use RestService.ForGenerated<T>(...) when the application should require a source-generated client at runtime. Otherwise, prefer RestService.For<T>(...) over reflection-heavy manual patterns around Type where possible.
  3. Supply source-generated System.Text.Json metadata for your DTOs.
  4. Do not reference the Refit.Reflection package. The reflection request builder is opt-in, so leaving it out keeps that pipeline out of your application entirely. Build with warnings as errors and the RF006 diagnostic will point at any method that still needs it.

For the default SystemTextJsonContentSerializer on .NET 8+, Refit prefers JsonTypeInfo metadata from your configured JsonSerializerOptions when it is available. That means Native AoT apps can improve compatibility by supplying source-generated metadata through a JsonSerializerContext or TypeInfoResolver on the serializer options they pass into SystemTextJsonContentSerializer.

[JsonSerializable(typeof(Todo))]
public partial class TodoJsonContext : JsonSerializerContext
{
}

var settings = new RefitSettings(
    new SystemTextJsonContentSerializer(
        new JsonSerializerOptions(JsonSerializerDefaults.Web)
        {
            TypeInfoResolver = TodoJsonContext.Default
        }
    )
);

var api = RestService.For<ITodoApi>("https://api.example.com", settings);

Generated-only client registration with IHttpClientFactory

AddRefitClient<T> is annotated with RequiresUnreferencedCode because it can fall back to the reflection-based registration path. For Native AoT or trimmed applications that rely on the source generator, use AddRefitGeneratedClient<T> (from Refit.HttpClientFactory) instead. It registers the client through IHttpClientFactory and RestService.ForGenerated<T>, so there is no reflection fallback and no RequiresUnreferencedCode warning, while the usual IHttpClientBuilder configuration (base address, message handlers, resilience pipelines) still applies:

services.AddRefitGeneratedClient<IWebApi>()
        .ConfigureHttpClient(c => c.BaseAddress = new Uri("https://api.example.com"));

// optionally with settings, a settings factory resolved from the container, or a custom HttpClient name
services.AddRefitGeneratedClient<IWebApi>(settings);
services.AddRefitGeneratedClient<IWebApi>(provider => new RefitSettings { /* configure settings */ });
services.AddRefitGeneratedClient<IWebApi>(settings, "my-named-client");

If no source-generated client exists for the interface, resolving it throws an InvalidOperationException that points you back to the generator output - the registration never silently falls back to reflection.

If a generated Refit client cannot be found at runtime, Refit now explicitly points you back to the source generator/build output and recommends generated clients plus source-generated System.Text.Json metadata for Native AoT scenarios.

Refit also ships analyzers for newer Roslyn toolchains, including a Roslyn 5.0 build for newer Visual Studio versions.

Handling exceptions

Refit has different exception handling behavior depending on if your Refit interface methods return Task<T> or if they return Task<IApiResponse>, Task<IApiResponse<T>>, or Task<ApiResponse<T>>.

When returning Task<IApiResponse>, Task<IApiResponse<T>>, or

Task<ApiResponse<T>>

Refit traps any HttpRequestException or TaskCanceledException raised by the HttpClient in an ApiRequestException. Refit also traps any ApiException raised by the ExceptionFactory when processing the response, and any errors that occur when attempting to deserialize the response to ApiResponse<T>. In both cases, it will populate the exception into the Error property on ApiResponse<T> without throwing the exception.

You can then decide what to do like so:

var response = await _myRefitClient.GetSomeStuff();
if(response.IsSuccessful)
{
   //do your thing
}
else
{
    // If you want to distinguish between request and response errors
    if (response.HasRequestError(out var requestError))
        _logger.LogError(requestError, "An error occurred while sending the request.");
    else if (response.HasResponseError(out var responseError))
        _logger.LogError(responseError, responseError.Content);

    // Or just log the error directly
    _logger.LogError(response.Error, "An error occurred while calling the API.");
}

Note

Migrating from v8-v11? response.Error.Content no longer compiles since v12. Error is now typed as ApiExceptionBase? (request-side context only), so the response body moved to the derived ApiException. Read it via response.HasResponseError(out var apiException) (as shown above) and then apiException.Content, or cast with (response.Error as ApiException)?.Content. See the V12.x.x breaking changes for the full before/after.

Note

The IsSuccessful property checks whether the response status code is in the range 200-299 and there wasn't any other error (for example, during content deserialization). If you just want to check the HTTP response status code, you can use the IsSuccessStatusCode property. Neither property implies a non-null Content — a successful response can have no body (e.g. 204 No Content or a literal null body). Use HasContent or IsSuccessfulWithContent when you need the deserialized Content to be present.

When returning Task<T>

Refit throws any exception raised by the HttpClient and wraps it in an ApiRequestException. It also throws any ApiException raised by the ExceptionFactory when processing the response and any errors that occur when attempting to deserialize the response to Task<T>.

// ...
try
{
   var result = await awesomeApi.GetFooAsync("bar");
}
catch (ApiRequestException exception)
{
    //exception handling for when a response was not received from the server
}
catch (ApiException exception)
{
   //exception handling for when a response was received from the server
}
// Or to not distinguish between request/response exceptions
catch (ApiExceptionBase exception)
{
   //exception handling for when an error occurs during the request/response
}
// ...

Refit can also throw ValidationApiException instead which in addition to the information present on ApiException also contains ProblemDetails when the service implements the RFC 7807 specification for problem details and the response content type is application/problem+json

For specific information on the problem details of the validation exception, simply catch ValidationApiException:

// ...
try
{
   var result = await awesomeApi.GetFooAsync("bar");
}
catch (ValidationApiException validationException)
{
   // handle validation here by using validationException.Content,
   // which is type of ProblemDetails according to RFC 7807

   // If the response contains additional properties on the problem details,
   // they will be added to the validationException.Content.Extensions collection.
}
catch (ApiException exception)
{
   // other exception handling
}
// ...

Inspecting the error body synchronously

ApiException exposes GetContentAsAsync<T>() to deserialize the error body, but it cannot be awaited inside an exception filter (catch ... when ...), which the CLR requires to be synchronous. Because the body is already buffered into the Content string, you can deserialize it synchronously with GetContentAs<T>() or the non-throwing TryGetContentAs<T>(out T? content). This lets you route to a specific handler only when the error body matches a well-known shape, and fall through to a generic handler otherwise:

try
{
    var result = await awesomeApi.GetFooAsync("bar");
}
catch (ApiException exception) when (exception.TryGetContentAs<Error>(out var error))
{
    // handle the strongly-typed `error` here
}
catch (ApiException exception)
{
    // generic handling when the body was missing or not an `Error`
}

TryGetContentAs<T> returns false (and never throws) when there is no content, the content does not deserialize, or the configured serializer cannot deserialize synchronously. GetContentAs<T> returns default when there is no content and throws NotSupportedException when the configured IHttpContentSerializer does not implement ISynchronousContentDeserializer. The built-in SystemTextJsonContentSerializer, NewtonsoftJsonContentSerializer, and XmlContentSerializer all implement it; a custom serializer can opt in by implementing ISynchronousContentDeserializer.

Reading the request body that was sent

By default HttpClient disposes the request content once a request is sent, so the body cannot be read back from ApiException.RequestMessage.Content. Set RefitSettings.CaptureRequestContent = true to buffer the request body into a string before sending and expose it on ApiExceptionBase.RequestContent:

var settings = new RefitSettings { CaptureRequestContent = true };
var awesomeApi = RestService.For<IAwesomeApi>("https://api.example.com", settings);

try
{
    await awesomeApi.PostFooAsync(payload);
}
catch (ApiException exception) when (exception.HasRequestContent)
{
    // exception.RequestContent holds exactly what Refit sent on the wire
    logger.LogError("Failed request body: {Body}", exception.RequestContent);
}

CaptureRequestContent defaults to false because it buffers the entire body into memory; avoid it for large or streamed uploads. When it is disabled, HasRequestContent is false and RequestContent is null.

Providing a custom ExceptionFactory

You can also override default exceptions behavior that are raised by the ExceptionFactory when processing the result by providing a custom exception factory in RefitSettings. For example, you can suppress all ApiExceptions with the following:

var nullTask = Task.FromResult<Exception>(null);

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com",
    new RefitSettings {
        ExceptionFactory = httpResponse => nullTask;
    });

For exceptions raised when attempting to deserialize the response use DeserializationExceptionFactory described bellow.

Providing a custom DeserializationExceptionFactory

You can override default deserialization exceptions behavior that are raised by the DeserializationExceptionFactory when processing the result by providing a custom exception factory in RefitSettings. For example, you can suppress all deserialization exceptions with the following:

var nullTask = Task.FromResult<Exception>(null);

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com",
    new RefitSettings {
        DeserializationExceptionFactory = (httpResponse, exception) => nullTask;
    });

Providing a custom TransportExceptionFactory

You can control the exception that is surfaced when HttpClient.SendAsync throws a transport-level error (for example, HttpRequestException, SocketException, or TaskCanceledException) by setting TransportExceptionFactory on RefitSettings. The factory receives the HttpRequestMessage that was being sent, the raw exception, and the CancellationToken that was passed to the call, and returns the exception that Refit will ultimately throw (non-ApiResponse path) or store in IApiResponse.Error (ApiResponse path).

By default, Refit wraps the transport exception in an ApiRequestException that carries the full request context, except when the exception is an OperationCanceledException and the CancellationToken is already cancelled — in that case the original exception is returned unchanged. You can replace that behavior entirely, for example to re-throw the original exception unchanged in all cases:

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com",
    new RefitSettings
    {
        TransportExceptionFactory = (request, ex, ct) => ex
    });

Because RefitSettings is captured in the closure, you can also reconstruct the default behavior with your own adjustments:

var settings = new RefitSettings();

settings.TransportExceptionFactory = (request, ex, ct) =>
{
    // Pass cancellation through so ApiResponse callers still throw rather
    // than capturing a cancelled request as a soft error.
    if (ex is OperationCanceledException && ct.IsCancellationRequested)
        return ex;

    // Wrap everything else in ApiRequestException, just like the default.
    return new ApiRequestException(request, request.Method, settings, ex);
};

var gitHubApi = RestService.For<IGitHubApi>("https://api.github.com", settings);

ApiException deconstruction with Serilog

For users of Serilog, you can enrich the logging of ApiException using the Serilog.Exceptions.Refit NuGet package. Details of how to integrate this package into your applications can be found here.

Testing your Refit clients

The first-party Refit.Testing package lets you test the Refit clients your app depends on without a mocking library or a live server. You describe the calls you expect as a route table — each entry pairs a Route (which request to match) with a Reply (what to send back) — and point a real Refit client at it:

using Refit.Testing;

var http = new StubHttp
{
    { Route.Get("/users/{id}"), Reply.With(new User(7, "octocat")) },
    { Route.Post("/users"),     Reply.Status(HttpStatusCode.Created) },
};

var api = http.CreateClient<IGitHubApi>("https://api.github.com");

var user = await api.GetUser(7);

// assert the body your client sent, as a typed object:
var sent = await http.LastRequestBodyAsync<NewUser>();
await http.VerifyAllCalledAsync();

Because the handler is Refit-aware, route templates mirror your [Get("/users/{id}")] attributes, Reply.With<T> serializes with the client's own serializer (no hand-written JSON), and you can read the sent request body back as a typed object. It also ships NetworkBehavior for seeded latency/fault injection and StubApiResponse<T> for unit-testing code that consumes IApiResponse<T> directly.

See the testing guide for the full walkthrough.

About

The automatic type-safe REST library for .NET Core, Xamarin and .NET. Heavily inspired by Square's Retrofit library, Refit turns your REST API into a live interface.

Topics

Resources

License

MIT, MIT licenses found

Licenses found

MIT
LICENSE
MIT
COPYING

Code of conduct

Stars

9.5k stars

Watchers

159 watching

Forks

Sponsor this project

  •  
  •  
  •  

Contributors