Fix it now
This is a network error wearing an identity code. Microsoft’s published remedy for 0xCAA82EE7 is to confirm the user has internet access and that the network permits the Microsoft endpoints Office needs – nothing about your licence is involved. Start by proving whether the name resolves and the port answers.
nslookup login.microsoftonline.com
netsh winhttp show proxy
- If
nslookupfails or returns an address belonging to your own filtering appliance, stop here: name resolution is the problem and nothing further matters yet. - If the name resolves, test the path:
Test-NetConnection login.microsoftonline.com -Port 443in PowerShell. - Read the
netsh winhttp show proxyoutput. Office background components use that machine-level configuration, not the browser’s. On a network where direct access is blocked, an empty machine proxy is enough to break sign-in on its own. - If it is empty and the browser works, test by importing the user’s settings:
netsh winhttp import proxy source=ie. Then retry sign-in. - Test on a different network, such as a phone hotspot. Working there confirms the fault is the original network rather than the machine.
netsh winhttp reset proxy puts the machine-level setting back to DIRECT if the import makes things worse. Deploy a real machine proxy configuration afterwards rather than leaving one imported by hand.
If Office signs in you can stop here. If it does not, the next section covers why Office and the browser can take different routes and which layer is actually failing.
Why it happens
Office and your browser do not necessarily reach the internet the same way. Browsers use the proxy settings configured for the signed-in user. Background components, the identity stack among them, frequently use the machine-level WinHTTP configuration instead, which is set separately with netsh and is often left empty on machines that were configured by hand. On a network that blocks direct access and permits only the proxy, that single difference breaks sign-in while everything else on the desktop appears healthy.
Microsoft’s published guidance for 0xCAA82EE7 is short and points squarely at the network: make sure the user has internet access, then verify that the network and the devices between the user and Microsoft are configured to allow the required addresses and ports. It lists the code alongside 0xCAA82EE2. That pairing is informative if you look at the low words – 0x2EE2 is WinINet 12002, the request timed out, and 0x2EE7 is WinINet 12007, the server name could not be resolved. Microsoft does not publish that decomposition, so treat it as a working hypothesis rather than a fact, but it lines up with the remedy Microsoft gives and it tells you where to point your first test.
The second common divergence is inspection. Many organisations terminate and re-issue TLS at the perimeter so traffic can be examined. Microsoft’s network connectivity guidance is explicit that Microsoft 365 domains should be bypassed from TLS decryption, traffic interception, deep packet inspection and content filtering, and lists TLS termination or deep packet inspection of Microsoft 365 domains among the practices known to cause connectivity problems. Applied to sign-in traffic it usually surfaces as a network-layer error rather than as a certificate warning, which is why it hides here.
The third is partial filtering. Sign-in is not one host: it involves the interactive sign-in endpoints, device registration and the Office licensing service, and a rule that permits some while blocking others produces exactly the intermittent behaviour people call random. Microsoft publishes the current list through a web service that firewalls and proxies can consume directly.
Name resolution is failing or being filtered
You have this one if nslookup on the sign-in endpoint returns nothing, or returns an address that belongs to your own filtering appliance.
- Confirm which resolvers the client is actually using:
ipconfig /all. - Compare against a known-good resolver to isolate the failure:
nslookup login.microsoftonline.com 1.1.1.1. - If your filtering product is answering for Microsoft names, permit them in its policy.
- Clear the client resolver cache with
ipconfig /flushdnsand retest.
The machine-level proxy is empty or wrong
You have this one if The browser reaches the internet without difficulty, Office cannot, and netsh winhttp show proxy reports no proxy on a network where direct access is not permitted.
- Test with
netsh winhttp import proxy source=ie, which Microsoft documents as the only source the import accepts. - Retry sign-in. If it now works, deploy the machine proxy configuration properly rather than leaving the import in place.
- To undo a bad configuration,
netsh winhttp reset proxyreturns the setting to DIRECT.
An automatic configuration script that returns direct access for Microsoft endpoints, on a network that does not permit direct access, produces exactly this symptom. Check what the script actually returns for the sign-in hosts rather than what it was meant to return.
Inspection is breaking the exchange
You have this one if Resolution and the port test both succeed, the proxy is configured, and sign-in still fails. Other Microsoft services may be intermittent too.
- Ask the network team whether Microsoft 365 domains are subject to break-and-inspect on this path.
- Bypass them from TLS decryption and deep packet inspection, which is what Microsoft’s connectivity guidance asks for.
- Retest from an affected client once the bypass is in place, not from the appliance.
Some endpoints are permitted and others are not
You have this one if Sign-in works for a while, or for some users, and the pattern does not correlate with anything on the client.
- Build the allow-list from Microsoft’s endpoints web service rather than from a static list, so it tracks the service.
- Match on hostname rather than IP address. These services sit behind content delivery networks and the addresses change without notice.
- Confirm the rules actually apply to the client subnets that are failing. That is usually where the gap turns out to be.
A captive portal is intercepting the request
You have this one if It fails only on guest or public networks, and a browser shows a sign-in page for the network itself.
- Complete the network’s own sign-in page in a browser, then retry Office sign-in.
- For managed devices that live on such networks, consider whether a VPN profile should establish before the applications start.
Full reference
Which test tells you what
| Test result | Where to look |
|---|---|
nslookup on the sign-in host fails |
DNS: an internal resolver, a filtering resolver, or a blocked forwarder |
| The name resolves but the port test fails | Firewall or routing between the client and the endpoint |
| Both succeed and Office still fails | Inspection, or the machine-level proxy configuration |
| Works on a hotspot, fails on the office network | The network, not the machine |
| Works for some users on one machine | A per-user proxy setting rather than a machine-level one |
| Fails only on a guest network | A captive portal that has not been completed |
The machine proxy commands, as documented
| Command | What it does |
|---|---|
netsh winhttp show proxy |
Displays the current WinHTTP proxy setting |
netsh winhttp set proxy <server>[:port] |
Sets the proxy server for the machine, with an optional bypass list |
netsh winhttp import proxy source=ie |
Imports the proxy settings from Internet Options. ie is the only source accepted |
netsh winhttp reset proxy |
Resets the WinHTTP proxy setting to DIRECT |
Building the allow-list without maintaining it by hand
Microsoft publishes the Microsoft 365 endpoints as a web service and recommends that firewalls, proxies and scripts consume it directly rather than reading a page. The identity endpoints – the set that includes login.microsoftonline.com – are published in the Allow category on TCP 443 and 80, together with the device registration and Entra endpoints that sign-in also touches. Building a rule from that service is the difference between an allow-list that ages and one that does not.
Do not build the rule from IP addresses. The published data changes, the services sit behind content delivery networks, and Microsoft’s own guidance is to work from the domain data in the web service.
What the codes in this family do and do not say
Of the four codes here, Microsoft publishes a meaning for one of them and a remedy for another. 0x80072EE6 is documented in the WinINet error list as ERROR_INTERNET_UNRECOGNIZED_SCHEME – the URL scheme could not be recognised or is not supported – which in practice means a malformed proxy configuration entry, and that is a real, checkable thing. 0xCAA82EE7 has a published remedy but no published definition. 0xCAA30194 and 0xCAA90019 have neither. Diagnose those two by the tests above rather than by anything a page tells you they mean.
When the network team says nothing has changed
- Ask for the gateway log for the client’s address at the exact time of the failure, not for a summary. A categorised block shows up immediately.
- Check whether the client subnet is covered by the same rules as the ones that work. Rules scoped to the wrong VLAN are the most common answer to “but it works for me”.
- Confirm whether an automatic configuration script is in play, and read what it returns for the sign-in hosts.
- Check for a second network path – a VPN, a mobile broadband adapter, a docking station’s NIC – that is being preferred over the one everyone assumes is in use.
- Test from a second machine on the same subnet before concluding anything about the first.
Why this is not a licensing problem
Nothing here touches entitlement. Office has not reached the point of asking who you are, let alone what you are allowed to run, so no seat, key or plan change alters the outcome. If somebody has suggested buying something to resolve a 0xCAA82EE7, the request never left the building and a purchase will not help it leave.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
0xCAA82EE7 |
Microsoft’s published remedy is to confirm internet access and that the network permits the required Microsoft endpoints and ports; no meaning is published beyond that | Microsoft Learn |
0xCAA30194 |
Seen in the same exchange during sign-in. Microsoft publishes no meaning for it | not published by the vendor |
0xCAA90019 |
Another network-layer failure from the same identity stack. Microsoft publishes no meaning for it | not published by the vendor |
0x80072EE6 |
WinINet 12006, ERROR_INTERNET_UNRECOGNIZED_SCHEME: the URL scheme could not be recognised, or is not supported | Microsoft Learn |
Confirm the fix worked
nslookup login.microsoftonline.comreturns Microsoft addresses rather than an internal appliance.Test-NetConnection login.microsoftonline.com -Port 443succeeds from the client subnet.netsh winhttp show proxyshows the configuration you intended the machine to have.- Office signs in and File, then Account shows the correct work account.
- A second machine on the same segment behaves the same way, which tells you the network was fixed rather than one client.
Questions people ask about this
Why does the browser work when Office cannot sign in?
They can use different proxy configurations. Browsers use the per-user settings; Office background components frequently use the machine-level WinHTTP configuration, which is set separately with netsh and is often empty on hand-built machines.
Can I just allow everything to Microsoft?
Broad ranges are neither necessary nor advisable, and IP-based rules go stale quickly. Microsoft publishes the endpoints as a web service that firewalls and proxies can consume directly, which keeps the list current without anyone maintaining it.
Is bypassing inspection for these endpoints a security risk?
It is a trade-off your security team should make knowingly, but bypassing Microsoft 365 domains from TLS decryption and deep packet inspection is Microsoft’s own published guidance, and it lists interception of those domains among the practices known to cause connectivity problems.
Does the last part of the code really mean DNS?
It lines up with WinINet 12007, the server name could not be resolved, and Microsoft lists this code next to 0xCAA82EE2, whose low word matches WinINet 12002, request timed out. Microsoft does not publish that mapping, so use it to choose your first test, not as a conclusion.
Does fixing this cost anything?
No. There is no licensing element to this error at all. Your subscription, key and entitlement are untouched, and every fix here is a network configuration change.
