Python Workers: ModuleNotFoundError: No module named 'workers' on deploy, even with official pywrangler init scaffold

[swiercnas-hash]

What versions & operating system are you using?

System:
OS: Windows 10 10.0.19045
CPU: Intel Core i5-4300U
Binaries:
Node: 24.18.0
npm: 11.16.0
npmPackages (worker project):
wrangler: 4.123.0
workers-py: 1.16.4
workers-runtime-sdk: 1.6.13

Please provide a link to a minimal reproduction

No hosted repo needed - reproduces with the official scaffold itself: uvx --from workers-py pywrangler init myapp --accept-defaults --no-deploy --no-git --no-open then cd myapp && uv run pywrangler deploy (unmodified generated entry.py/pyproject.toml/wrangler.jsonc).

Describe the Bug

Deploying any Python Worker fails at the Cloudflare API validation step with a ModuleNotFoundError for the 'workers' package - even the entry.py generated by the official scaffold.

Steps to reproduce:

1. uvx --from workers-py pywrangler init myapp --accept-defaults --no-deploy --no-git --no-open
2. 2. cd myapp
3. 3. uv run pywrangler deploy
      Expected: deploy succeeds with the unmodified generated template.
      Actual: fails with the error pasted below. The traceback shows it originates from Cloudflare's own internal introspection.py trying from workers import (...), not from user code - the generated entry.py's own from workers import Response, WorkerEntrypoint line is where it dies.

I also tried a hand-written minimal Worker (single file, async def on_fetch(request, env) + js.Response.new(...), zero dependencies, wrangler.toml with only compatibility_flags = ["python_workers"]) via plain wrangler deploy - identical error.

Things I tried that did not fix it:

• compatibility_flags: ["python_workers", "disable_python_external_sdk"] - gets past this specific error, but then fails differently with "The uploaded script has no registered event handlers" [code: 10068].
• • Upgrading wrangler 4.112.0 -> 4.123.0 - no change.
• • • Empty requirements.txt, and requirements.txt containing just "workers" - no change (this was before I found the pywrangler/pyproject.toml flow; same error persists with that flow too).
      Account is on the Workers Free plan, if relevant.

Please provide any relevant error logs

X [ERROR] A request to the Cloudflare API (/accounts/.../workers/scripts/myapp/versions) failed.

Uncaught Error: PythonError: Traceback (most recent call last):
File "/lib/python313.zip/_pyodide/_base.py", line 523, in eval_code
.run(globals, locals)
~~~^^^^^^^^^^^^^^^^^
File "/lib/python313.zip/_pyodide/_base.py", line 357, in run
coroutine = eval(self.code, globals, locals)
File "", line 9, in
File "/lib/python313.zip/_pyodide/_base.py", line 666, in pyimport_impl
res = import(stem, fromlist=fromlist)
File "/session/metadata/entry.py", line 10, in
from workers import Response, WorkerEntrypoint
File "/lib/python3.13/site-packages/workers/init.py", line 3, in
raise err
ModuleNotFoundError: No module named 'workers'
You need to update to workers-py >= 1.90 or to pass disable_python_external_sdk

at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:19919:14) in new_error
at [object Object] in $wrap_exception
at [object Object] in $pythonexc2js
at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:22411:37) in callPyObjectKwargs
at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:23309:20) in callKwargs
at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:24469:87) in runPython
at null.<anonymous> (pyodide:python-entrypoint-helper:73:17) in handleSrcImport
at null.<anonymous> (pyodide:python-entrypoint-helper:190:17)

[code: 10021]

[workers-devprod]

Note

This is an automated, advisory triage report generated by workers-devprod. It is not an official maintainer response — a maintainer will follow up.

🤖 Automated triage report

Issue Triage: #212

Summary

Deploying any Python Worker — including the unmodified official pywrangler init scaffold — fails at the Cloudflare API validation/introspection step with ModuleNotFoundError: No module named 'workers'.

Classification

• Type: Bug
• Component: packages/wrangler (Python Workers deploy path; python label). Root cause likely in the workers-py / workers-runtime-sdk packaging (repo: cloudflare/workers-py).
• Severity: High
• Has Reproduction: Yes (self-contained steps against the official scaffold; no external repo needed)
• Quality: Complete

Findings

• Created: 2026-08-15
• Author: swiercnas-hash
• Version: wrangler 4.123.0, workers-py 1.16.4, workers-runtime-sdk 1.6.13 (Node 24.18.0, npm 11.16.0, Windows 10)
• Labels: python
• Comments: 4 (2 automated triage bot comments, 1 detailed reporter follow-up, 1 confirmation from a second user)

Analysis

• Reproducible and well-scoped. The reporter gives exact copy-pasteable steps using the official scaffold with unmodified generated files (uvx --from workers-py pywrangler init ... && uv run pywrangler deploy). They also reproduced it with a hand-written zero-dependency Worker via plain wrangler deploy — identical error. The failure is in the scaffold's own entry.py line from workers import Response, WorkerEntrypoint, before any user code.
• Independently confirmed. A second user (waylan) reports the identical error and traceback on Windows 11 for a brand-new project, also on workers-py==1.16.4. This rules out a one-off account/state problem and points to a real, shared defect.
• Error message is self-contradictory / suggests a version-parse bug. The runtime error says "You need to update to workers-py >= 1.90", yet both affected users are on 1.16.4. 1.16.4 is newer than 1.90 under semver, so a naive string/float comparison of the version (treating 1.16 as < 1.9) would explain why the guard fires incorrectly. waylan explicitly notes Version comparison bug: workers-py >= 1.90 requirement blocks upgrade from 1.9.3 #120 covered only the 1.9.0 vs 1.90 message typo, not the underlying logic.
• No clean workaround. Adding disable_python_external_sdk gets past the ModuleNotFoundError but then fails with "The uploaded script has no registered event handlers" [code: 10068]. Deleting and redeploying a fresh script, empty requirements.txt, and requirements.txt containing workers all reproduce the same error. Upgrading wrangler 4.112.0 → 4.123.0 made no difference.
• Triage bot could not reproduce — but only on a temporary account. The auto-triage repro succeeded via --temporary (unauthenticated throwaway account). The reporter shows --temporary is rejected once authenticated, and hits the error every time on their real Free-plan account. The temporary-vs-authenticated difference may be a meaningful variable (e.g. plan tier, SDK-injection path) and should be checked when reproducing.

Recommendation

Status: NEEDS VERIFICATION

Reasoning: This is a complete, independently-confirmed bug that blocks the officially documented Python Workers deploy path, with no clean workaround and no fix from upgrading. The one automated repro that "passed" used a temporary account, so it does not disprove the report; a maintainer needs to reproduce on a real authenticated account and pinpoint whether the fix belongs in wrangler's Python deploy/introspection path or in the workers-py / workers-runtime-sdk packaging.

Action: Route to a maintainer with Python Workers context. Reproduce on a real authenticated account (not --temporary), ideally on the Workers Free plan and on Windows, following the reporter's exact steps. Investigate the workers-py >= 1.90 version check for a string/float version-comparison bug that misfires for 1.16.x. If the root cause is in the SDK packaging, transfer to / open a companion issue in cloudflare/workers-py; keep this open until confirmed.

Suggested Labels: python

Suggested Comment

Thanks for the very thorough write-up, and for the follow-up timeline — reproducing this against the unmodified pywrangler init scaffold, plus the second confirmation from @waylan on Windows 11, is really helpful.

One detail stands out: the error says "update to workers-py >= 1.90", but you're both on 1.16.4, which is already newer than 1.90. That looks like it could be a version-comparison bug (treating 1.16 as lower than 1.9), so the SDK guard fires when it shouldn't. We'll dig into that.

A couple of things that would help us narrow it down:

• Can either of you reproduce on macOS or Linux, or does it only happen on Windows?
• Does it still fail if you pin workers-py to an earlier version (e.g. 1.9.x)?

Note that the earlier automated repro succeeded on a temporary (unauthenticated) account, which behaves differently from an authenticated one, so it doesn't rule out what you're seeing. We'll reproduce on a real account and follow up. If it turns out the fix belongs in the Python SDK, we'll track it in cloudflare/workers-py.

Structured summary (JSON)

{
  "issueNumber": 15208,
  "title": "Python Workers: ModuleNotFoundError: No module named 'workers' on deploy, even with official pywrangler init scaffold",
  "githubUrl": "https://github.com/cloudflare/workers-py/issues/212",
  "recommendation": "NEEDS VERIFICATION",
  "difficulty": "medium",
  "reasoning": "Complete, independently-confirmed bug blocking the official Python Workers deploy path, with no clean workaround and unaffected by upgrading; the only 'passing' repro used a temporary account so it needs verification on a real authenticated account. The 'update to workers-py >= 1.90' message misfiring for 1.16.4 suggests a version-comparison bug.",
  "suggestedAction": "Route to a Python Workers maintainer to reproduce on a real authenticated account (not --temporary), investigate the workers-py >= 1.90 version-comparison guard for a string/float bug misfiring on 1.16.x, and determine whether the fix belongs in wrangler's deploy path or in the workers-py/workers-runtime-sdk packaging (cloudflare/workers-py).",
  "hasSuggestedComment": true,
  "suggestedLabels": ["python"]
}

[workers-devprod]

• Reproduction: ❌ Could not reproduce. Official scaffold with workers-py 1.16.4 deployed successfully to a temporary Cloudflare account using uv run pywrangler deploy --temporary. Worker responded correctly to requests, rather than failing with the reported ModuleNotFoundError during API validation.

[swiercnas-hash]

Thanks for the triage! A couple of things worth noting since this happens consistently and I've been able to narrow it down a bit further:

Timeline of what I tried, in order, all on the same real (non-temporary) authenticated account, Workers Free plan:

1. Hand-written single-file Worker (async def on_fetch(request, env), zero deps, plain wrangler deploy) -> ModuleNotFoundError.
2. 2. Same, with compatibility_flags: ["python_workers", "disable_python_external_sdk"] added -> gets past the import, fails differently with "no registered event handlers" [code: 10068].
3. 3. Reverted to just ["python_workers"], added empty requirements.txt -> same ModuleNotFoundError as wrangler dev workers-sdk#1.
4. 4. requirements.txt containing just workers -> same ModuleNotFoundError as wrangler dev workers-sdk#1.
5. 5. Switched to the official flow: uvx --from workers-py pywrangler init scaffold, pyproject.toml with workers-py/workers-runtime-sdk dev deps, class-based Default(WorkerEntrypoint), deployed via uv run pywrangler deploy -> same ModuleNotFoundError, now failing on the scaffold's own entry.py line from workers import Response, WorkerEntrypoint (not even my code yet).
6. 6. uv run pywrangler deploy --temporary -> rejected: "You're already authenticated with Cloudflare, so --temporary can't be used."
7. 7. Deleted the Worker entirely (wrangler delete) and deployed fresh (brand new script, no prior deploy history) -> identical ModuleNotFoundError, same traceback.
      So it's not stale/corrupted state on a specific Worker - a brand new script hits it too, on this account, every time, regardless of which of the two documented workarounds I try. Given the bot's repro attempt succeeded on a temporary (unauthenticated, throwaway) account rather than a real authenticated one, and step 6 shows --temporary is unavailable once authenticated, I can't currently produce a passing case on my own account to compare against. If it'd help, happy to share the account ID privately or run any diagnostic command a maintainer suggests.

[waylan]

I'm getting the same error in the same circumstances. The only difference is that I am on Windows 11 rather than WIndows 10. This is a new install for a new project and deploy always results in the same error.

X [ERROR] service core:user:dingus: Uncaught Error: PythonError: Traceback (most recent call last):

    File "/lib/python313.zip/_pyodide/_base.py", line 523, in eval_code
      .run(globals, locals)
       ~~~^^^^^^^^^^^^^^^^^
    File "/lib/python313.zip/_pyodide/_base.py", line 357, in run
      coroutine = eval(self.code, globals, locals)
    File "<exec>", line 9, in <module>
    File "/lib/python313.zip/_pyodide/_base.py", line 666, in pyimport_impl
      res = __import__(stem, fromlist=fromlist)
    File "/session/metadata/entry.py", line 1, in <module>
      from workers import Response, WorkerEntrypoint
    File "/lib/python3.13/site-packages/workers/__init__.py", line 3, in <module>
      raise err
  ModuleNotFoundError: No module named 'workers'
  You need to update to workers-py >= 1.90 or to pass disable_python_external_sdk

    at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:19919:14) in new_error
    at [object Object] in $wrap_exception
    at [object Object] in $pythonexc2js
    at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:22411:37) in callPyObjectKwargs
    at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:23309:20) in callKwargs
    at null.<anonymous> (pyodideRuntime-internal:emscriptenSetup:24469:87) in runPython
    at null.<anonymous> (pyodide:python-entrypoint-helper:73:17) in handleSrcImport
    at null.<anonymous> (pyodide:python-entrypoint-helper:190:17)

For completeness, I did find #120, but that seems to be about a typo in the error message (1.9.0 instead of 1.90), not the code. As I and the original reporter both have workers-py==1.16.4, we should be well beyond that issue.