[Bug]: Docker v0.9.0,cant using dashboard,ERR_EMPTY_RESPONSE

[AlexTuring32]

crawl4ai version

0.9.0

Expected Behavior

just wanna use crawl4ai

Current Behavior

windows 11 pro/wsl2/docker desktop/
docker pull unclecode/crawl4ai:latest
docker run -d -p 11235:11235 --name crawl4ai --shm-size=1g unclecode/crawl4ai:latest
http://localhost:11235/dashboard
return:ERR_EMPTY_RESPONSE

i try to change port
docker run -d -p 5001:11235 --name crawl4ai --shm-size=3g unclecode/crawl4ai:latest
but still cant use

and than.
i try to add -e CRAWL4AI_API_TOKEN="xxxxxxxxxx"
docker run -d -p 5001:11235 --name crawl4ai -e CRAWL4AI_API_TOKEN="xxxxxxxxx" --shm-size=3g unclecode/crawl4ai:latest
it works~
return: "detail": "Authentication required"

Is this reproducible?

Yes

Inputs Causing the Bug

Steps to Reproduce

Code snippets

OS

win 11 pro

Python version

3.14

Browser

chrome

Browser version

No response

Error logs & Screenshots (if applicable)

PS C:\WINDOWS\system32> docker run -d -p 5001:11235 --name crawl4ai -e CRAWL4AI_API_TOKEN="xxxxxxxxxx" --shm-size=3g unclecode/crawl4ai:latest
034960e7d7440d4fb3ad37782eadd90d37200c183f6341c13a43b3c94d2163c6
PS C:\WINDOWS\system32> docker logs crawl4ai
entrypoint: no REDIS_PASSWORD provided; generated an ephemeral one.
2026-06-25 07:29:52,449 CRIT could not write pidfile /app/supervisord.pid
2026-06-25 07:29:53,452 INFO spawned: 'redis' with pid 8
2026-06-25 07:29:53,454 INFO spawned: 'gunicorn' with pid 9
8:C 25 Jun 2026 07:29:53.462 * oO0OoO0OoO0Oo Redis is starting oO0OoO0OoO0Oo
8:C 25 Jun 2026 07:29:53.462 * Redis version=8.8.0, bits=64, commit=00000000, modified=1, pid=8, just started
8:C 25 Jun 2026 07:29:53.462 * Configuration loaded
8:M 25 Jun 2026 07:29:53.463 * monotonic clock: POSIX clock_gettime
8:M 25 Jun 2026 07:29:53.463 * Running mode=standalone, port=6379.
8:M 25 Jun 2026 07:29:53.464 * Server initialized
8:M 25 Jun 2026 07:29:53.464 * Ready to accept connections tcp
[2026-06-25 07:29:53 +0000] [9] [INFO] Starting gunicorn 26.0.0
[2026-06-25 07:29:53 +0000] [9] [INFO] Listening at: http://[::]:11235 (9)
[2026-06-25 07:29:53 +0000] [9] [INFO] Using worker: uvicorn.workers.UvicornWorker
[2026-06-25 07:29:53 +0000] [15] [INFO] Booting worker with pid: 15
[2026-06-25 07:29:53 +0000] [9] [INFO] Control socket listening at /home/appuser/.gunicorn/gunicorn.ctl
2026-06-25 07:29:54,898 INFO success: redis entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
2026-06-25 07:29:54,898 INFO success: gunicorn entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
No SECRET_KEY set. Auto-generated an ephemeral key (changes on restart, invalidating issued tokens). Set SECRET_KEY for any real deployment.
/app/job.py:49: UserWarning: Field name "schema" in "LlmJobPayload" shadows an attribute in parent "BaseModel"
class LlmJobPayload(BaseModel):
/app/server.py:1012: FastAPIDeprecationWarning: regex has been deprecated, please use pattern instead
context_type: str = Query("all", regex="^(code|doc|all)$"),
MCP server running on 127.0.0.1:11235
[2026-06-25 07:29:55 +0000] [15] [INFO] Started server process [15]
[2026-06-25 07:29:55 +0000] [15] [INFO] Waiting for application startup.
2026-06-25 07:29:55,595 - server - INFO - Auth gate active (credential configured).
2026-06-25 07:29:55,601 - crawl4ai.egress - INFO - egress pinning proxy listening on http://127.0.0.1:37907
2026-06-25 07:29:55,602 - crawl4ai.workqueue - INFO - work queue started (maxsize=1000, workers=4, per_principal=unlimited)
2026-06-25 07:29:55,604 - monitor - INFO - Started persistence worker
2026-06-25 07:29:55,633 - crawler_pool - INFO - 🔥 Creating permanent default browser
[INIT].... → Crawl4AI 0.9.0
[2026-06-25 07:29:56 +0000] [15] [INFO] Application startup complete.
PS C:\WINDOWS\system32> curl http://localhost:5001/health

安全警告: 脚本执行风险
Invoke-WebRequest 可解析网页内容。解析网页时，可能会运行网页中的脚本代码。
建议的操作:
使用 -UseBasicParsing 开关来避免执行脚本代码。

  是否要继续?

[Y] 是(Y) [A] 全是(A) [N] 否(N) [L] 全否(L) [S] 暂停(S) [?] 帮助 (默认值为“N”): Y

StatusCode : 200
StatusDescription : OK
Content : {"status":"ok","timestamp":1782372691.02042,"version":"0.9.0"}
RawContent : HTTP/1.1 200 OK
x-content-type-options: nosniff
x-frame-options: DENY
referrer-policy: no-referrer
cross-origin-opener-policy: same-origin
cross-origin-resource-policy: same-origin
content-secur...
Forms : {}
Headers : {[x-content-type-options, nosniff], [x-frame-options, DENY], [referrer-policy, no-referrer], [cross
-origin-opener-policy, same-origin]...}
Images : {}
InputFields : {}
Links : {}
ParsedHtml : mshtml.HTMLDocumentClass
RawContentLength : 62

PS C:\WINDOWS\system32> docker run -d -p 5001:11235 --name crawl4ai --shm-size=3g unclecode/crawl4ai:latest
d0d52c6f38758152f7c27b4b3e0ce024fc8073dc2054436dc5ae7e2cd84a2742
PS C:\WINDOWS\system32> curl http://localhost:5001/health
curl : 基础连接已经关闭: 连接被意外关闭。
所在位置 行:1 字符: 1

• curl http://localhost:5001/health

•   + CategoryInfo          : InvalidOperation: (System.Net.HttpWebRequest:HttpWebRequest) [Invoke-WebRequest]，WebExce
    ption
    + FullyQualifiedErrorId : WebCmdletWebResponseException,Microsoft.PowerShell.Commands.InvokeWebRequestCommand
  
  

PS C:\WINDOWS\system32> docker logs crawl4ai
entrypoint: no REDIS_PASSWORD provided; generated an ephemeral one.
entrypoint: no CRAWL4AI_API_TOKEN set; binding loopback only (127.0.0.1:11235).
2026-06-25 07:33:33,480 CRIT could not write pidfile /app/supervisord.pid
2026-06-25 07:33:34,484 INFO spawned: 'redis' with pid 8
2026-06-25 07:33:34,486 INFO spawned: 'gunicorn' with pid 9
8:C 25 Jun 2026 07:33:34.494 * oO0OoO0OoO0Oo Redis is starting oO0OoO0OoO0Oo
8:C 25 Jun 2026 07:33:34.494 * Redis version=8.8.0, bits=64, commit=00000000, modified=1, pid=8, just started
8:C 25 Jun 2026 07:33:34.494 * Configuration loaded
8:M 25 Jun 2026 07:33:34.494 * monotonic clock: POSIX clock_gettime
8:M 25 Jun 2026 07:33:34.494 * Running mode=standalone, port=6379.
8:M 25 Jun 2026 07:33:34.495 * Server initialized
8:M 25 Jun 2026 07:33:34.495 * Ready to accept connections tcp
[2026-06-25 07:33:34 +0000] [9] [INFO] Starting gunicorn 26.0.0
[2026-06-25 07:33:34 +0000] [9] [INFO] Listening at: http://127.0.0.1:11235 (9)
[2026-06-25 07:33:34 +0000] [9] [INFO] Using worker: uvicorn.workers.UvicornWorker
[2026-06-25 07:33:34 +0000] [15] [INFO] Booting worker with pid: 15
[2026-06-25 07:33:34 +0000] [9] [INFO] Control socket listening at /home/appuser/.gunicorn/gunicorn.ctl
2026-06-25 07:33:35,989 INFO success: redis entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
2026-06-25 07:33:35,989 INFO success: gunicorn entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
No SECRET_KEY set. Auto-generated an ephemeral key (changes on restart, invalidating issued tokens). Set SECRET_KEY for any real deployment.
/app/job.py:49: UserWarning: Field name "schema" in "LlmJobPayload" shadows an attribute in parent "BaseModel"
class LlmJobPayload(BaseModel):
2026-06-25 07:33:36,565 - crawl4ai.security - WARNING - CRAWL4AI_API_TOKEN is not set. All API endpoints are unauthenticated. Set CRAWL4AI_API_TOKEN environment variable to enable authentication.
/app/server.py:1012: FastAPIDeprecationWarning: regex has been deprecated, please use pattern instead
context_type: str = Query("all", regex="^(code|doc|all)$"),
MCP server running on 127.0.0.1:11235
[2026-06-25 07:33:36 +0000] [15] [INFO] Started server process [15]
[2026-06-25 07:33:36 +0000] [15] [INFO] Waiting for application startup.
2026-06-25 07:33:36,605 - server - WARNING - No CRAWL4AI_API_TOKEN set; generated an ephemeral token for this loopback session: CRAWL4AI_API_TOKEN=5b0ef31beb97e31a8a2804371574a68c187ac09f1c82b5863ccd182cc76446f1
2026-06-25 07:33:36,611 - crawl4ai.egress - INFO - egress pinning proxy listening on http://127.0.0.1:33393
2026-06-25 07:33:36,612 - crawl4ai.workqueue - INFO - work queue started (maxsize=1000, workers=4, per_principal=unlimited)
2026-06-25 07:33:36,615 - monitor - INFO - Started persistence worker
2026-06-25 07:33:36,644 - crawler_pool - INFO - 🔥 Creating permanent default browser
[INIT].... → Crawl4AI 0.9.0
[2026-06-25 07:33:37 +0000] [15] [INFO] Application startup complete.

[ssh-manu]

I had the same issue. Without a CRAWL4AI_API_TOKEN crawl4ai only listen on localhost inside the docker, which in fact you can't access. To use the dashboard and playground you need to set Authorization: Bearer xxxxxxxxx header, which is not possible in a browser. Dashboard and playground don't accept ?token=xxxxxxxxx URL param. I think dashboard and playground are broken in Docker + 0.9.0.

[sevenmoonlightsteps]

Dug into this on current develop. There are two separate things in the report; the second is the real UI bug.

1. No token set -> ERR_EMPTY_RESPONSE. This is the intended 0.9.0 secure-by-default behavior, not a UI bug. With no CRAWL4AI_API_TOKEN (and JWT off), _resolve_auth() / the entrypoint bind the server to loopback only, so the port isn't reachable from outside the container and the browser gets an empty response. Setting a token (or enabling JWT) makes it bind normally.

2. Token set -> {"detail":"Authentication required"} when opening /dashboard or /playground. This is the actual bug, and it's more fundamental than the UI just "forgetting" a header.

AuthGateMiddleware (deploy/docker/auth_gate.py) is the outermost ASGI gate and covers every route and static mount. Its public allowlist is only:

# deploy/docker/server.py
app.add_middleware(
    AuthGateMiddleware,
    token_provider=_current_api_token,
    public_paths={HEALTH_PATH, "/token"},
)

So /dashboard, /playground, /static/* and /monitor/* all require a credential. Two consequences:

• For HTTP, the gate only reads the token from the Authorization: Bearer header. The ?token= query-param fallback is gated to WebSocket connections only (auth_gate.py:94-99). A browser can't attach an Authorization header to a top-level navigation, so just opening http://host:11235/dashboard in the address bar is rejected with 401 before any HTML is served. The page can never bootstrap.
• Even if it did load, the static pages send no credential on their calls: static/playground/index.html and static/monitor/index.html issue plain fetch() / new WebSocket() with no Authorization header and no ?token=, so every data call would 401 / close 4401.

(_UI_PREFIXES in server.py is only a CSP exemption for the inline-script pages, not an auth exemption, so it doesn't help here.)

Suggested fix (keeps the API fail-closed):

1. Let the static UI shell load without a credential, since it serves no data, only HTML/CSS/JS. Either add the UI mounts to public_paths (e.g. /playground, /dashboard, /static) or mount them ahead of the gate.
2. Give the UI a way to acquire and carry a token: a small token field (or a login that posts to the existing /token endpoint), persisted in sessionStorage, then attach Authorization: Bearer <token> to every fetch() and append ?token=<token> to the monitor WebSocket URL (the gate already accepts ?token= on WS, and monitor_routes.py references it).

That keeps every data/API route gated while letting the dashboard and playground actually authenticate from a browser.

Happy to open a PR along these lines if that direction looks right. Wanted to confirm the approach first since it touches the new auth boundary and I didn't want to second-guess the secure-by-default design.

[ntohidi]

Root cause: AuthGateMiddleware (deploy/docker/auth_gate.py) is the outermost ASGI gate and covers every route. Its public_paths allowlist only includes /health and /token — so /dashboard, /playground, and /static/* all require a credential.

Two problems make the UI unreachable:

1. HTTP token extraction only accepts Authorization: Bearer header (auth_gate.py:87-93). The ?token= query param fallback is WebSocket-only (auth_gate.py:95-99). Browsers can't attach custom headers to top-level navigation, so opening http://host:11235/dashboard in the address bar is rejected with 401 before any HTML is served.
2. The static UI pages send no credentials on their API calls — all fetch() calls in playground/index.html and monitor/index.html have no Authorization header.

Fix (in progress): Let the static UI shell (HTML/CSS/JS) load without credentials via public_prefixes on the middleware — it serves no data. Then add a token input bar to both UIs that stores the token in sessionStorage and attaches Authorization: Bearer <token> to all fetch() and ?token= to WebSocket calls. All data/API routes stay fail-closed.