Undici ships with a set of built-in interceptors that can be composed via dispatcher.compose() to add cross-cutting behaviour such as automatic retries, response decompression, redirect following, DNS caching, and more.

import { Agent, interceptors } from 'undici'

const { retry, redirect, decompress, dump, responseError, dns, cache, deduplicate } = interceptors

const agent = new Agent().compose([
  retry({ maxRetries: 3 }),
  redirect({ maxRedirections: 5 }),
  decompress()
])

const response = await agent.request({ origin: 'https://example.com', path: '/', method: 'GET' })

You can also apply interceptors to a single Client or Pool:

import { Client, interceptors } from 'undici'

const client = new Client('https://example.com').compose(
  interceptors.retry({ maxRetries: 2 })
)

Custom interceptors use the same shape as dispatcher.compose(): an interceptor takes a dispatch function and returns another dispatch-like function with the same (options, handler) signature.

When an interceptor wraps the handler, forward the callbacks that it does not handle itself. The complete handler callback list is documented under dispatcher.dispatch(options, handler).

import { Agent } from 'undici'

const timingInterceptor = dispatch => {
  return (options, handler) => {
    const started = performance.now()

    return dispatch(options, {
      ...handler,
      onResponseStart (controller, statusCode, headers, statusMessage) {
        const duration = Math.round(performance.now() - started)
        const method = options.method ?? 'GET'
        const origin = options.origin ?? ''

        console.log(`${method} ${origin}${options.path} -> ${statusCode} in ${duration}ms`)

        return handler.onResponseStart?.(
          controller,
          statusCode,
          headers,
          statusMessage
        )
      },
      onResponseError (controller, error) {
        const duration = Math.round(performance.now() - started)

        console.error(`request failed after ${duration}ms`, error)

        return handler.onResponseError?.(controller, error)
      }
    })
  }
}

const dispatcher = new Agent().compose(timingInterceptor)

const { body } = await dispatcher.request({
  origin: 'https://example.com',
  path: '/',
  method: 'GET'
})

await body.dump()
await dispatcher.close()

Reads and discards the response body up to a configurable size limit. Useful for keeping a connection alive after an error response without reading the body yourself.

Parameters

• opts {Object} (optional)
  • maxSize {number} Maximum number of bytes to read and discard. Responses whose declared or received body size exceeds this value are aborted. Default: 1_048_576 (1 MiB).

Per-request override: set dumpMaxSize on the dispatch options to override the global maxSize for a specific request.

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors } from 'undici'

const agent = new Agent().compose(
  interceptors.dump({ maxSize: 128 * 1024 }) // discard up to 128 KiB
)

Automatically retries failed requests using the same options accepted by RetryHandler.

Parameters

• opts {RetryHandler.RetryOptions} (optional) Global retry options applied to every request. Individual requests can override via opts.retryOptions. See RetryOptions for the full list of accepted fields.

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors } from 'undici'

const agent = new Agent().compose(
  interceptors.retry({
    maxRetries: 5,
    minTimeout: 200,
    maxTimeout: 5000,
    timeoutFactor: 2,
    statusCodes: [429, 502, 503, 504]
  })
)

Follows HTTP redirects (3xx responses) automatically.

Parameters

• opts {Object} (optional)
  • maxRedirections {number} Maximum number of redirects to follow. Passing 0 disables redirect following entirely. Default: undefined (inherits from the per-request maxRedirections option).
  • throwOnMaxRedirect {boolean} When true, throws an error when a redirect would be followed but the redirect limit has been reached, instead of returning that redirect response. A non-redirect response after exactly maxRedirections redirects is returned normally. Default: false.
  • stripHeadersOnRedirect {string[]} List of header names to remove from the request when following any redirect. Default: [].
  • stripHeadersOnCrossOriginRedirect {string[]} List of header names to remove from the request when following a cross-origin redirect (i.e. the redirect target has a different origin). Useful for stripping Authorization on cross-origin hops. Default: [].

Per-request override: any of the four options above can also be set directly on the dispatch options to override the interceptor defaults for a specific request.

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors } from 'undici'

const agent = new Agent().compose(
  interceptors.redirect({
    maxRedirections: 10,
    throwOnMaxRedirect: true,
    stripHeadersOnCrossOriginRedirect: ['authorization', 'cookie']
  })
)

Automatically decompresses response bodies encoded with gzip, x-gzip, br (Brotli), deflate, or zstd. Any other coding, including compress and x-compress, is left untouched, matching fetch(). As required by RFC 9659, a zstd frame that requires a window size larger than 8 MB fails to decode.

Experimental: This interceptor is experimental and subject to change. A one-time ExperimentalWarning is emitted on first use.

Parameters

• opts {Object} (optional)
  • skipStatusCodes {number[]} Status codes for which decompression is skipped. Default: [204, 304].
  • skipErrorResponses {boolean} When true, responses with a status code

    = 400 are not decompressed. Default: true.

  • maxSize {number} Maximum decompressed response size in bytes for each decompression stage. The request fails with a ResponseExceededMaxSizeError if a stage exceeds this limit. Set to 0 to disable the limit. Default: 0.

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors } from 'undici'

const agent = new Agent().compose(
  interceptors.decompress({
    skipStatusCodes: [204, 304],
    skipErrorResponses: false, // decompress error bodies too
    maxSize: 16 * 1024 * 1024 // limit decoded bodies to 16 MiB
  })
)

Converts 4xx/5xx responses into thrown ResponseError instances, making it easy to handle HTTP errors with a standard try/catch block.

The error body is automatically decoded for application/json and text/plain responses. For JSON responses the body is parsed and exposed as error.body.

Parameters

• opts {Object} (optional)
  • maxSize {number} Maximum number of response body bytes to include in the error. If the body exceeds this limit, the resulting ResponseError contains the retained prefix with bodyTruncated set to true. Set to 0 to disable the limit. Default: 1048576 (1 MiB).

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors, errors } from 'undici'

const agent = new Agent().compose(interceptors.responseError())

try {
  await agent.request({ origin: 'https://example.com', path: '/not-found', method: 'GET' })
} catch (err) {
  if (err instanceof errors.ResponseError) {
    console.error(err.status, err.body)
  }
}

Caches DNS lookups so that repeated requests to the same origin reuse the resolved IP address instead of performing a fresh lookup every time. Supports dual-stack (IPv4 + IPv6) and custom lookup/storage implementations.

Parameters

• opts {Object} (optional)
  • maxTTL {number} Maximum number of milliseconds a DNS record is cached, regardless of the TTL returned by the resolver. Default: 0 (use the TTL from the DNS record).
  • maxItems {number} Maximum number of origins to cache simultaneously. Oldest entries are evicted when the limit is reached. Default: Infinity.
  • dualStack {boolean} When true, both IPv4 (A) and IPv6 (AAAA) records are looked up and the interceptor picks between them based on affinity. Default: true.
  • affinity {4 | 6 | null} Preferred IP family when dualStack is enabled. null lets the interceptor alternate between families. Default: null.
  • lookup {Function} (optional) Custom DNS resolution function with the same signature as node:dns's lookup callback form: (origin, options, callback) => void.
  • pick {Function} (optional) Custom record-selection function called with (origin, records, affinity) to choose which resolved address to use.
  • storage {DNSStorage} (optional) Custom storage backend. Must implement get, set, delete, full, and size.

DNSStorage interface

interface DNSStorage {
  size: number
  get(origin: string): DNSInterceptorOriginRecords | null
  set(origin: string, records: DNSInterceptorOriginRecords | null, options: { ttl: number }): void
  delete(origin: string): void
  full(): boolean
}

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors } from 'undici'

const agent = new Agent().compose(
  interceptors.dns({
    maxTTL: 60_000, // cache for at most 60 seconds
    dualStack: true,
    affinity: 4     // prefer IPv4
  })
)

Caches HTTP responses according to RFC 9111 (HTTP Caching). See CacheStore for information on providing a custom backing store.

Parameters

• opts {CacheHandler.CacheOptions} (optional) See the CacheStore documentation for accepted fields.

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors, cacheStores } from 'undici'

const agent = new Agent().compose(
  interceptors.cache({ store: new cacheStores.MemoryCacheStore() })
)

Deduplicates concurrent identical requests so that only one is sent over the wire. All waiting callers receive the same response once the in-flight request completes. Only safe HTTP methods (e.g. GET, HEAD) may be deduplicated.

Parameters

• opts {Object} (optional)
  • methods {string[]} HTTP methods to deduplicate. Must be safe HTTP methods (GET, HEAD, OPTIONS, TRACE). Default: ['GET'].
  • skipHeaderNames {string[]} Header names whose presence in a request causes it to bypass deduplication entirely. Matching is case-insensitive. Default: [].
  • excludeHeaderNames {string[]} Header names to exclude from the deduplication key. Requests that differ only in these headers are still deduplicated together. Useful for headers like x-request-id that vary per request but should not prevent deduplication. Matching is case-insensitive. Default: [].
  • maxBufferSize {number} Maximum number of bytes buffered per paused waiting handler. If a waiting handler exceeds this threshold it is failed with an AbortError to prevent unbounded memory growth. Default: 5_242_880 (5 MiB).

Returns: {Dispatcher.DispatcherComposeInterceptor}

Example

import { Agent, interceptors } from 'undici'

const agent = new Agent().compose(
  interceptors.deduplicate({
    methods: ['GET', 'HEAD'],
    excludeHeaderNames: ['x-request-id', 'x-trace-id'],
    maxBufferSize: 2 * 1024 * 1024
  })
)

Interceptors are applied in the order they appear in the compose() call. The first interceptor in the array wraps the outermost layer.

import { Agent, interceptors } from 'undici'

const agent = new Agent().compose([
  interceptors.dns({ maxTTL: 30_000 }),
  interceptors.retry({ maxRetries: 3 }),
  interceptors.redirect({ maxRedirections: 5 }),
  interceptors.decompress(),
  interceptors.responseError()
])

In the example above the request flow is:

1. dns — resolves and caches the target IP
2. retry — retries the dispatch on transient failures
3. redirect — follows any 3xx redirects
4. decompress — decompresses the response body
5. responseError — converts 4xx/5xx into thrown errors