Minor Changes

• #13726 b5ac54b Thanks @penalosa! - Hard fail on Node.js < 22

  Wrangler no longer supports Node.js 20.x, as it reached end-of-life on 2026-04-30. The minimum supported Node.js version is now 22.0.0. See https://github.com/nodejs/release?tab=readme-ov-file#end-of-life-releases.

• #13717 9a1f014 Thanks @NuroDev! - Add an experimental experimental_generateTypes() programmatic API.

  Wrangler now exposes experimental_generateTypes() from the package root so you can generate Worker types in code using the same logic as wrangler types. The API supports the same core type-generation options (include env/runtime toggles) and returns structured output with separate env and runtime content alongside the combined formatted output.

Patch Changes

• #13732 22e1a61 Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  workerd	1.20260426.1	1.20260429.1

• #13754 00523c8 Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  workerd	1.20260429.1	1.20260430.1

• #13711 1c4d850 Thanks @dario-piotrowicz! - fix: skip auto-config and OpenNext delegation when --config is explicitly provided

  When --config is passed to wrangler deploy, the user is explicitly targeting a specific Worker configuration. Previously, wrangler would ignore --config and delegate to opennextjs-cloudflare deploy if it detected an OpenNext project in the working directory, silently deploying the wrong Worker. Now, both auto-config detection and OpenNext delegation are skipped when --config is provided, matching the existing behavior for --script and --assets.

• #13735 6d28037 Thanks @edmundhung! - Improve config-schema.json hover text in more editors

  Wrangler now emits markdownDescription in config-schema.json alongside the existing description field. Editors that support rich JSON Schema hovers can use that markdown directly instead of rendering escaped links and formatting.

• #13722 0827815 Thanks @MattieTK! - Improve safe telemetry categorisation for user-facing Wrangler errors.

• #13116 e539008 Thanks @dario-piotrowicz! - Allow getPlatformProxy and unstable_getMiniflareWorkerOptions to start when the assets directory does not exist yet

  Previously, getPlatformProxy would catch and swallow NonExistentAssetsDirError internally when the configured assets directory was absent on disk. This has been refactored so that the directory-existence check is skipped entirely for getPlatformProxy and unstable_getMiniflareWorkerOptions, since these APIs are typically used at dev time in frameworks where the assets directory is a build output that may not exist yet.

  wrangler dev, wrangler deploy, wrangler versions upload, and wrangler triggers deploy continue to require the assets directory to exist when specified.

• Updated dependencies [22e1a61, 00523c8, b5ac54b, e653edf, e1eff94, e539008, 0bf64a7, b04eedf, 6457fb3, c07d0cb]:

  • miniflare@4.20260430.0
  • @cloudflare/kv-asset-handler@0.5.0

Minor Changes

• #13726 b5ac54b Thanks @penalosa! - Hard fail on Node.js < 22

  Wrangler no longer supports Node.js 20.x, as it reached end-of-life on 2026-04-30. The minimum supported Node.js version is now 22.0.0. See https://github.com/nodejs/release?tab=readme-ov-file#end-of-life-releases.

• #13390 0bf64a7 Thanks @Ltadrian! - Fix Hyperdrive binding issue where some customers are unable to connect to local databases using wrangler dev

  • Skips creating a local TCP proxy server for Hyperdrive bindings when SSL is not enabled, connecting directly to the database instead. This avoids connection refused errors caused by firewall rules or proxy port binding issues on Windows/macOS.

• #13565 b04eedf Thanks @vaishnav-mk! - Add restart from step support for local Workflows development

  Workflow instances can now be restarted from a specific step in local development. When restarting from a step, all earlier steps preserve their cached results and replay instantly, while the target step and everything after it re-execute.

  The WorkflowInstance.restart() method now accepts an optional { from: { name, count?, type? } } parameter to specify which step to restart from.

• #13618 c07d0cb Thanks @jamesopstad! - Support V2 protocol for module fallback service

  When the new_module_registry compatibility flag is set, requests sent to unsafeModuleFallbackService() use a different protocol. Miniflare now supports both protocols and exports a parseModuleFallbackRequest() utility to ease handling.

Patch Changes

• #13732 22e1a61 Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  workerd	1.20260426.1	1.20260429.1

• #13754 00523c8 Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  workerd	1.20260429.1	1.20260430.1

• #13723 e653edf Thanks @edmundhung! - Expose send_email bindings from getPlatformProxy()

  Projects developing in Node can now access send_email bindings from the platform proxy. This supports the plain-object MessageBuilder API locally, so calls like env.EMAIL.send({ from, to, subject, text }) no longer fail because the binding is missing.

• #12514 e1eff94 Thanks @ascorbic! - fix: normalise typed array subclasses in devalue serialization

  Node.js Buffer extends Uint8Array but isn't available in all runtimes. When a Buffer was passed through the proxy serialization bridge (e.g. as a D1 bind parameter via getPlatformProxy()), the reviver would fail because "Buffer" isn't in the allowed constructor list and may not exist on globalThis in workerd.

  The reducer now normalises subclass constructor names to the nearest standard typed array parent before serialization, matching structured clone behaviour.

• #13116 e539008 Thanks @dario-piotrowicz! - Gracefully handle a missing assets directory by starting with zero assets

  Previously, configuring Miniflare with an assets.directory that did not exist on disk would cause the asset services to fail to start. This is a common situation during wrangler dev when the assets directory is a build output that hasn't been generated yet.

  Now, when the configured assets directory does not exist, Miniflare creates an empty temporary directory and starts the asset services with zero assets. Once the real directory is created and setOptions() is called (e.g. triggered by the file watcher), Miniflare reloads and begins serving the actual assets.

• #13363 6457fb3 Thanks @courtney-sims! - Prepares router-worker for a more gradual rollout by refactoring and separating out the invocation from the business logic. In the future, this will provide space for us to route requests to new versions of router-worker based on their plan, but should make no functional difference today.

Minor Changes

• #13726 b5ac54b Thanks @penalosa! - Hard fail on Node.js < 22

  Wrangler no longer supports Node.js 20.x, as it reached end-of-life on 2026-04-30. The minimum supported Node.js version is now 22.0.0. See https://github.com/nodejs/release?tab=readme-ov-file#end-of-life-releases.

Patch Changes

• #13359 fc2d883 Thanks @dependabot! - Update dependencies of "create-cloudflare"

  The following dependency versions have been updated:

  Dependency	From	To
  @angular/create	21.2.6	21.2.8

• #13729 83a1c7e Thanks @dependabot! - Update dependencies of "create-cloudflare"

  The following dependency versions have been updated:

  Dependency	From	To
  nuxi	3.34.0	3.35.1

• #13730 c5bcdfa Thanks @dependabot! - Update dependencies of "create-cloudflare"

  The following dependency versions have been updated:

  Dependency	From	To
  @tanstack/create-start	0.59.22	0.59.26

• #13731 4d4d2c2 Thanks @dependabot! - Update dependencies of "create-cloudflare"

  The following dependency versions have been updated:

  Dependency	From	To
  sv	0.15.1	0.15.2

• #13733 60ee337 Thanks @dependabot! - Update dependencies of "create-cloudflare"

  The following dependency versions have been updated:

  Dependency	From	To
  create-analog	2.4.10	2.5.0

Minor Changes

• #13565 b04eedf Thanks @vaishnav-mk! - Add restart from step support for local Workflows development

  Workflow instances can now be restarted from a specific step in local development. When restarting from a step, all earlier steps preserve their cached results and replay instantly, while the target step and everything after it re-execute.

  The WorkflowInstance.restart() method now accepts an optional { from: { name, count?, type? } } parameter to specify which step to restart from.

Patch Changes

• #13363 6457fb3 Thanks @courtney-sims! - Prepares router-worker for a more gradual rollout by refactoring and separating out the invocation from the business logic. In the future, this will provide space for us to route requests to new versions of router-worker based on their plan, but should make no functional difference today.

Patch Changes

• Updated dependencies [22e1a61, 00523c8, b5ac54b, e653edf, e1eff94, 1c4d850, 6d28037, 9a1f014, e539008, 0bf64a7, 0827815, b04eedf, 6457fb3, c07d0cb, e539008]:
  • miniflare@4.20260430.0
  • wrangler@4.87.0

Minor Changes

• #13618 c07d0cb Thanks @jamesopstad! - Support V2 protocol for module fallback service

  When the new_module_registry compatibility flag is set, requests sent to unsafeModuleFallbackService() use a different protocol. The Vite plugin now supports both protocols in its handling of additional module types.

Patch Changes

• #13363 6457fb3 Thanks @courtney-sims! - Prepares router-worker for a more gradual rollout by refactoring and separating out the invocation from the business logic. In the future, this will provide space for us to route requests to new versions of router-worker based on their plan, but should make no functional difference today.

• Updated dependencies [22e1a61, 00523c8, b5ac54b, e653edf, e1eff94, 1c4d850, 6d28037, 9a1f014, e539008, 0bf64a7, 0827815, b04eedf, 6457fb3, c07d0cb, e539008]:

  • miniflare@4.20260430.0
  • wrangler@4.87.0

Patch Changes

• Updated dependencies [22e1a61, 00523c8, b5ac54b, e653edf, e1eff94, e539008, 0bf64a7, b04eedf, 6457fb3, c07d0cb]:
  • miniflare@4.20260430.0

Minor Changes

• #13726 b5ac54b Thanks @penalosa! - Hard fail on Node.js < 22

  Wrangler no longer supports Node.js 20.x, as it reached end-of-life on 2026-04-30. The minimum supported Node.js version is now 22.0.0. See https://github.com/nodejs/release?tab=readme-ov-file#end-of-life-releases.

Minor Changes

• #13605 ea943ff Thanks @danielgek! - Add namespace support to wrangler ai-search commands

  All wrangler ai-search instance commands (create, list, get, update, delete, stats, search) now accept a --namespace (or -n) flag to target a specific AI Search namespace. When the flag is omitted, commands default to the default namespace that Cloudflare automatically provisions for every account.

  wrangler ai-search list now displays a namespace column, and wrangler ai-search create offers an interactive picker for existing namespaces (with an option to create a new one) when --namespace is not supplied in an interactive session.

  A new wrangler ai-search namespace subcommand group is also introduced, with list, create, get, update, and delete subcommands for managing namespaces directly.

  wrangler ai-search list --namespace blog
  wrangler ai-search create my-instance --namespace blog --type r2 --source my-bucket
  wrangler ai-search namespace create blog --description "Blog content"

• #13637 9eb9e69 Thanks @edmundhung! - Add --tunnel flag to wrangler dev for sharing your local dev server via Cloudflare Quick Tunnels

  You can now expose your local dev server publicly by passing --tunnel:

  wrangler dev --tunnel

  This starts a Cloudflare Quick Tunnel that gives you a random *.trycloudflare.com URL to share. The tunnel stops automatically when the dev session ends. Quick tunnels don't require a Cloudflare account or any configuration.

  A warning is shown when Server-Sent Events (SSE) responses are detected through the tunnel, since quick tunnels don't support SSE.

• #13661 0a5db08 Thanks @aspizu! - wrangler tail will now log stack traces. These stack traces already include resolved frames if you have chosen to upload sourcemaps.

• #13617 118027d Thanks @roerohan! - Force Flagship bindings to always use remote mode in local dev

  Flagship bindings now always access the remote Flagship service during local development, matching the behavior of AI bindings. Previously, Flagship supported both local and remote modes, but the local stub only returned default values, providing no real functionality and creating a dual source of truth for flag evaluations.

  The remote config field is retained for backward compatibility but only controls whether a warning is displayed. Setting remote: true suppresses the warning that Flagship bindings always access remote resources and may incur usage charges in local dev.

• #13254 e867ac2 Thanks @tgarg-cf! - Add wrangler queues consumer list subcommands for listing queue consumers

  Three new commands are available for listing consumers on a queue:

  • wrangler queues consumer list <queue-name> — lists all consumers (both worker and HTTP pull), grouped by type
  • wrangler queues consumer worker list <queue-name> — lists only worker consumers
  • wrangler queues consumer http list <queue-name> — lists only HTTP pull consumers

Patch Changes

• #13696 62e9f2a Thanks @dependabot! - Update dependencies of "miniflare", "wrangler"

  The following dependency versions have been updated:

  Dependency	From	To
  workerd	1.20260424.1	1.20260426.1

• #13576 2dc6175 Thanks @MattieTK! - Restore telemetry tracking for common CLI flags that were unintentionally dropped during sanitisation

  When argument sanitisation was introduced, only explicitly allow-listed args had their values included in telemetry. The allow list was very conservative, which meant common boolean flags like --remote, --json, --dry-run, --force, and many others were no longer being captured in sanitizedArgs despite previously being tracked. Boolean flags are inherently safe (values are only true/false), so these have now been added back to the global allow list. A small number of fixed-choice args (--local-protocol, --upstream-protocol, --containers-rollout) have also been added with their known value sets.

• #13649 ae8eae3 Thanks @petebacondarwin! - Fix service binding and tail consumer props being dropped between workers in different local dev instances

  When a service binding or tail consumer configured with props targeted a worker running in a separate wrangler dev instance (via the dev registry), the props were silently dropped and the remote entrypoint saw an empty ctx.props. Props are now forwarded correctly across the dev registry boundary, matching the behavior users get when all workers run in a single instance.

  // wrangler.json
  {
    "services": [
      {
        "binding": "AUTH",
        "service": "auth-worker", // may be in a separate `wrangler dev` process
        "entrypoint": "SessionEntry",
        "props": { "tenant": "acme" }
      }
    ]
  }

  The target worker's SessionEntry entrypoint now correctly receives { tenant: "acme" } on ctx.props regardless of which local dev instance it runs in.

• #13662 f2e2241 Thanks @petebacondarwin! - Fix three resource leaks in unstable_startWorker teardown that could prevent Node from exiting cleanly after worker.dispose().

  • The esbuild context created by bundleWorker is now disposed when the initial build fails. Previously a failing initial build (e.g. an unresolvable entrypoint, or a worker started with an invalid config via setConfig) left the esbuild child process running for the lifetime of the parent Node process.
  • runBuild's cleanup function now awaits the in-flight build before running the bundler's stop handler. Previously teardown could return before esbuild.BuildContext.dispose() had been called, so the esbuild watcher kept the event loop alive after dispose had resolved.
  • BundlerController.teardown() now runs the esbuild cleanup before removing the bundler's temporary directory, and aborts the in-flight bundle build so it cannot emit stale bundleStart/bundleComplete events after teardown. Previously the tmpdir was removed first, which in race with an in-flight rebuild produced confusing "Could not resolve .wrangler/tmp/bundle-XXXX/middleware-loader.entry.ts" errors during dispose.

• #13674 4f6ed93 Thanks @petebacondarwin! - Stop emitting a misleading [wrangler:error] Docker build exited with code: <n> log when the user aborts an in-progress container image build (for example by pressing the r rebuild hotkey while the previous build is still running).

  The abort-detection branch in the local and multi-worker runtime controllers was matching the wrong error message — it checked for "Build exited with code: 1", but the error thrown by the docker build helper is actually "Docker build exited with code: <n>", and the exit code after a process-group SIGINT/SIGKILL is typically 130/137/143, not 1. As a result, every legitimate user-initiated rebuild abort produced a spurious error event and [wrangler:error] log line. The check now matches the real error message prefix and ignores any non-zero exit code from the aborted build, so a user-requested rebuild while another build is in progress is silent.

• #13667 ed2f4ec Thanks @emily-shen! - fix: Preserve auth in remote proxy session data to avoid unnecessary session restarts

  maybeStartOrUpdateRemoteProxySession was not including auth in its return value, so on subsequent calls preExistingRemoteProxySessionData.auth was always undefined. This caused the auth comparison to always detect a change, disposing and recreating the remote proxy session on every reload even when auth had not changed.

• #13695 92bb8a5 Than...