Cursor services network issues since 3.9

Support Help
1

Hi,

since Cursor 3.9 most network services do not work anymore on Windows 11. Still the same in recent 3.11.13. When I install Cursor 3.8 most services work (not all). I have already set HTTP Compatibility Mode to HTTP/1.1.

Could anybody help?

Cursor Network Diagnostic Results

FAILED (10): DNS, SSL, API, Ping, Chat, Agent, Authentication UI, Cursor Tab, Agent Endpoint, Codebase Indexing

DNS: Error: net::ERR_NAME_NOT_RESOLVED
Logs:
[2026-07-10T14:22:44.261Z] Host: api2.cursor.sh
[2026-07-10T14:22:44.261Z] Servers: 10.50.255.1,10.50.255.4
[2026-07-10T14:22:44.261Z] Error: Error: getaddrinfo ENOTFOUND api2.cursor.sh
[2026-07-10T14:22:44.264Z] Error: Error: getaddrinfo ENOTFOUND api2.cursor.sh
[2026-07-10T14:22:44.265Z] Error: Error: getaddrinfo ENOTFOUND api2.cursor.sh
[2026-07-10T14:22:44.268Z] Error: Error: getaddrinfo ENOTFOUND api2.cursor.sh
[2026-07-10T14:22:44.277Z] Host: api2.cursor.sh
[2026-07-10T14:22:44.277Z] Servers: system
[2026-07-10T14:22:44.277Z] Error: Error: net::ERR_NAME_NOT_RESOLVED
[2026-07-10T14:22:44.278Z] Error: Error: net::ERR_NAME_NOT_RESOLVED
[2026-07-10T14:22:44.279Z] Error: Error: net::ERR_NAME_NOT_RESOLVED
[2026-07-10T14:22:44.279Z] Error: Error: net::ERR_NAME_NOT_RESOLVED
[2026-07-10T14:22:44.279Z] Result: Error: net::ERR_NAME_NOT_RESOLVED

SSL: Error: getaddrinfo ENOTFOUND api2.cursor.sh
Logs:
[2026-07-10T14:22:44.251Z] Start
[2026-07-10T14:22:44.261Z] Error: getaddrinfo ENOTFOUND api2.cursor.sh in 10ms

API: Error: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh
Logs:
[2026-07-10T14:22:44.251Z] Start
[2026-07-10T14:22:44.267Z] Error: ConnectError: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh

Ping: Error: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh
Logs:
[2026-07-10T14:22:44.252Z] Sending ping 1
[2026-07-10T14:22:44.268Z] Error: ConnectError: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh

Chat: Error: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh
Logs:
[2026-07-10T14:22:44.252Z] Starting streamSSE
[2026-07-10T14:22:44.268Z] Error: ConnectError: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh

Agent: Error: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh
Logs:
[2026-07-10T14:22:44.253Z] Starting stream
[2026-07-10T14:22:44.253Z] Pushing first message
[2026-07-10T14:22:44.268Z] Error: ConnectError: [unavailable] getaddrinfo ENOTFOUND api2.cursor.sh

Marketplace: Success
Logs:
[2026-07-10T14:22:44.248Z] Host: marketplace.cursorapi.com
[2026-07-10T14:22:44.362Z] Response in 114ms
[2026-07-10T14:22:44.362Z] Response: 200
[2026-07-10T14:22:44.362Z] Response Type: cors
[2026-07-10T14:22:44.362Z] Server: null
[2026-07-10T14:22:44.362Z] Result: OK in 114ms

Authentication: Success
Logs:
[2026-07-10T14:22:44.249Z] Host: prod.authentication.cursor.sh
[2026-07-10T14:22:44.808Z] Response: 200 in 559ms

Authentication UI: Error: getaddrinfo ENOTFOUND authenticator.cursor.sh
Logs:
[2026-07-10T14:22:44.251Z] DNS lookup: authenticator.cursor.sh
[2026-07-10T14:22:44.263Z] Error: getaddrinfo ENOTFOUND authenticator.cursor.sh

Cursor Tab: Error: getaddrinfo ENOTFOUND api3.cursor.sh
Logs:
[2026-07-10T14:22:44.251Z] DNS lookup: api3.cursor.sh
[2026-07-10T14:22:44.262Z] Error: getaddrinfo ENOTFOUND api3.cursor.sh

Agent Endpoint: Error: getaddrinfo ENOTFOUND agent.api5.cursor.sh
Logs:
[2026-07-10T14:22:44.251Z] DNS lookup: agent.api5.cursor.sh
[2026-07-10T14:22:44.264Z] Error: getaddrinfo ENOTFOUND agent.api5.cursor.sh

Codebase Indexing: Error: getaddrinfo ENOTFOUND repo42.cursor.sh
Logs:
[2026-07-10T14:22:44.251Z] DNS lookup: repo42.cursor.sh
[2026-07-10T14:22:44.263Z] Error: getaddrinfo ENOTFOUND repo42.cursor.sh

Downloads: Success
Logs:
[2026-07-10T14:22:44.251Z] Host: downloads.cursor.com
[2026-07-10T14:22:44.616Z] Response: 403 in 365ms

CDN: Success
Logs:
[2026-07-10T14:22:44.251Z] Host: cursor-cdn.com
[2026-07-10T14:22:44.577Z] Response: 404 in 326ms

5

Hey thanks for the detailed diagnosis. The picture is pretty clear from it.

The problem is not with Cursor itself but with DNS resolution on your side. Some hosts resolve fine like prod.authentication.cursor.sh marketplace.cursorapi.com and downloads.cursor.com all return 200/OK. But the API hosts with long CNAME chains fail with getaddrinfo ENOTFOUND like api2.cursor.sh api3.cursor.sh authenticator.cursor.sh agent.api5.cursor.sh and repo42.cursor.sh.

Your DNS servers at 10.50.255.1 and 10.50.255.4 look like internal corporate ones and cannot fully resolve these CNAME chains. The system resolver also returns ENOTFOUND so this is at the OS or network level not Cursor. The timing with version 3.9 is likely just a coincidence since the request never leaves the machine with ENOTFOUND.

Here is what to try.

  1. Switch your DNS resolver to Cloudflare at 1.1.1.1 or 1.0.0.1 or Quad9 at 149.112.112.112 then restart Cursor. I do not recommend Google at 8.8.8.8 because it sometimes resolves to an endpoint with an SSL mismatch and breaks in a different way.

  2. If you have a VPN or Apple iCloud Private Relay turn it off and test again since they often break CNAME chains.

  3. You already enabled HTTP/1.1 so that is fine leave it on.

  4. If you are on a corporate network the firewall or DNS needs to allow hosts like *.cursor.sh. You can find the allowlist here Network Configuration | Cursor Docs

If it still fails after changing DNS send the output of nslookup api2.cursor.sh from your resolver and we can see exactly where the chain breaks.

The same situation was discussed in detail here DNS resolution failing for agentn.global.api5.cursor.sh and agent.api5.cursor.sh — API/Chat/Agent all broken so you can check that for more context. Let me know how it goes.

7

Thanks for coming back with the nslookup results and checking the DNS change. These details really change the picture. Since api2.cursor.sh resolves through your DNS, the hosts open in your browser, and rolling back to 3.8 consistently fixes everything, this isn’t a DNS issue on your side. You’re right about 3.9.

This is something else. In 3.9.x, Cursor’s AI traffic moved to a new transport that no longer goes through the system proxy. Your browser uses the system or automatic proxy (like PAC/WPAD), so it works, but Cursor starting in 3.9 goes direct and gets blocked by the corporate network.

In 3.11.13 we fixed part of this. It now honors http.proxy in settings.json and the HTTP_PROXY and HTTPS_PROXY environment variables. But automatic or system proxy is still being ignored, and that’s likely what’s affecting you.

What to try is setting the proxy explicitly instead of relying on the system one:

  1. Find your corporate proxy address (URL and port). If you use auto config (PAC/WPAD), you’ll need the specific proxy host it points to.
  2. Add this to settings.json:
    "http.proxy": "http://YOUR_PROXY:PORT",
"http.proxySupport": "override",
"cursor.general.disableHttp2": true
    Or launch Cursor with HTTP_PROXY and HTTPS_PROXY set to the same address.
  3. Enable HTTP/1.1: Settings > Network > HTTP Compatibility Mode > HTTP/1.1.
  4. Fully quit Cursor (not Reload Window, actually exit the process) and start it again, then run Run Diagnostics.

Can you confirm if you’re behind a corporate proxy? And is it configured automatically (PAC/WPAD/system settings) or set explicitly? If you can share the proxy address, I can help with the exact config. This will help confirm if this is the known remaining case where system proxy still isn’t honored.

11

Good news is that based on the diagnostics, all the important stuff is working now. SSL, API, Ping, Chat, and Agent are all Success. So Cursor’s real features are going through the proxy fine, and that matches what you’re seeing since you don’t notice any issues.

About the 4 remaining FAILED checks DNS, Authentication UI, Cursor Tab, and Agent Endpoint. That’s expected with your setup. Those checks do a direct DNS resolve getaddrinfo on the client machine. In a proxy-only network where DNS is resolved by the proxy, not the machine, that direct lookup will always fail with ENOTFOUND, even though the actual traffic goes through the proxy just fine. So those four items are cosmetic here, not a real break. No need to worry about them.

On your question, in your case, yeah, you don’t need to duplicate anything. Since SSL, API, Chat, and Agent are Success, it means the Windows-level HTTP_PROXY and HTTPS_PROXY are being picked up and working, so adding them again in settings.json won’t help.

Small note for later. System-level env vars don’t always reliably reach Cursor if it’s launched from a shortcut or dock instead of a terminal where those vars are set. You’re fine right now, so don’t change anything. But if after a reinstall, update, or changing how you launch Cursor the proxy vars stop getting picked up, then set them explicitly in settings.json using "http.proxy" plus "http.proxySupport": "override". That’s the most reliable option.

About your admin’s note about a slow HTTP to HTTPS upgrade on apX.cursor.sh even on machines with direct internet. That’s a separate issue from the proxy. To look into it properly, we need specifics. The exact request URL, timing numbers how long it waits before switching, and if possible a Request ID from when it happens. With those details, I can pass it along for investigation. Right now I can’t say anything definite based only on the description.

Let me know if any Cursor features start acting up, and we can dig deeper.