Cursor publishes per-component status at status.cursor.com covering IDE, CLI, Cloud Agents, Automations, Bugbot, and cursor.com. Many incidents there are degradations at the model providers rather than Cursor itself, so check the provider status page too. If everything is green, run Cursor Settings then Network then Run Diagnostics, and switch HTTP Compatibility Mode to HTTP/1.1 if you are behind a corporate proxy or VPN.
- Official status: status.cursor.com, broken out into six components.
- A large share of incidents are model provider degradations, not Cursor.
- Downdetector is a crowd signal, not a source. Use it to corroborate.
- Green status and no connection usually means HTTP/2 through a proxy.
- Subscribe to the Atom or RSS feed so you find out before you waste ten minutes.
Check it in thirty seconds
In order, because each step rules out the one below it.
1. https://status.cursor.com official, per-component
2. the status page of the model you have selected:
https://status.anthropic.com Claude models
https://status.openai.com GPT models
https://status.cloud.google.com Gemini models
and the vendor page for anything else you have pinned
3. downdetector.com, search Cursor crowd reports, corroboration onlyWhat each answer means.
| What you see | It means | Do this |
|---|---|---|
| Cursor status shows an incident | It is not you | Read the affected component, then wait |
| Cursor green, provider degraded | Very common | Switch to a different model |
| Both green, others reporting on Downdetector | Something regional or partial | Check your network, then wait |
| Both green, nobody else reporting | It is your machine or network | Skip to the local fixes |
The six components, and which one you care about
The status page is not a single up or down light. It reports six components separately, and most of the time when people say "Cursor is down" only one of them is affected. Knowing which one you are using saves you from waiting out an incident that has nothing to do with your work.
Components on status.cursor.com, August 2026.
| Component | What it covers | You notice when |
|---|---|---|
| IDE | The desktop editor and its AI features | Chat, agent, or Tab stops responding |
| Cloud Agents | Agents that run on Cursor infrastructure | A background task never starts or never finishes |
| Automations | Scheduled and triggered agent runs | A rule fires and nothing happens |
| CLI | The command-line agent | Terminal runs fail while the editor is fine |
| Bugbot | Automated code review on pull requests | PR reviews stop appearing |
| cursor.com | Website, dashboard, login, billing | You cannot sign in or see usage |
Most Cursor incidents are not Cursor
Cursor routes your requests to model providers. When one of those providers has a bad hour, Cursor logs it as an incident on its own status page, because from your seat it is indistinguishable from Cursor breaking. Read the incident titles and this becomes obvious quickly: a large share of them name a model or a vendor rather than a Cursor system.
The two families of incident, and what each means for you.
| Family | Typical title | Your move |
|---|---|---|
| Upstream model | Degraded performance for a named model or vendor | Switch model. Often fixes it instantly. |
| Cursor system | Cloud Agents, Automations, or a widespread issue | Nothing to switch to. Wait it out. |
Read which model the incident names
If it names a vendor or a specific model, the fix is in your hands. If it names a Cursor subsystem, it is not.
Switch to a model from a different vendor
Not merely a different model from the same one. If Anthropic models are degraded, another Anthropic model is likely degraded too.
Or drop to Auto
Auto mode lets Cursor route to something it believes is healthy, and it does not draw on your credit pool. During a single-vendor incident this is usually the fastest way back to working.
"Taking longer than expected"
This exact string is its own search, which tells you how often it happens. The agent accepts your prompt, the spinner runs, and the message appears instead of a reply. It is not one bug; it is four different failures that surface identically.
Causes in rough order of likelihood.
| Cause | Tell | Fix |
|---|---|---|
| Upstream model degraded | Status page names that vendor | Switch model, or drop to Auto |
| HTTP/2 blocked by a proxy or VPN | Corporate network, works on your phone hotspot | HTTP Compatibility Mode to HTTP/1.1 |
| Stale build | It started after an update or has been weeks | Command palette, Cursor: Attempt Update |
| A very large agent turn | Huge context, many files, Max mode | Narrow the request; name the files |
Change model first
Ten seconds, and it resolves the most common cause. If a reply comes back on a different vendor's model, you have your answer and you can stop.
Test the network somewhere else
Tether to your phone for one prompt. If it works on the hotspot and not on the office network, it is HTTP/2 through a proxy and the next step fixes it permanently.
Switch to HTTP/1.1
Cursor Settings, then Network, then set HTTP Compatibility Mode to HTTP/1.1. Cursor uses HTTP/2 for streaming and some proxies do not handle bidirectional streams; this switches to a shape they do.
Update, then restart properly
Command palette with Cmd or Ctrl plus Shift plus P, run Cursor: Attempt Update, and restart when prompted. Quit fully rather than closing the window, since background processes can survive a window close.
Run the built-in diagnostics
Cursor Settings, then Network, then Run Diagnostics. It separates DNS, certificate, proxy, and protocol failures, which is a much better starting point than guessing.
Slow, but not down
Slow and down have different causes and the fix depends entirely on what is slow. Separate these before troubleshooting anything.
| What is slow | Likely cause | Fix |
|---|---|---|
| Replies take much longer than usual | Upstream provider under load | Switch model; check the provider status page |
| Tab completions lag | Local resources or a huge file | Close other editors, split the file |
| The whole editor is sluggish | Extensions, a very large workspace, low disk | Disable extensions, run Clear Editor History |
| Only agent runs are slow | Large context or Max mode | Name the files; reserve Max mode for genuinely hard work |
| It got slow after an update | Bad build for your setup | Update again, or switch release channel in settings |
Green status, still not working
If both status pages are clean, the problem is between your machine and Cursor. In practice it is nearly always one of three things, and all three have documented fixes.
1. HTTP/2 through a proxy or VPN
Cursor uses HTTP/2 for its streaming AI features. Some enterprise proxies and VPNs do not handle bidirectional HTTP/2 streams, and the symptom is that the editor works fine while every AI feature hangs. Cursor Settings, then Network, then set HTTP Compatibility Mode to HTTP/1.1.
2. Domains blocked by a firewall
On a managed network these need to be reachable:
*.cursor.sh includes authenticate.cursor.sh and authenticator.cursor.sh
*.cursor-cdn.com
*.cursorapi.com includes marketplace.cursorapi.com3. Signed out without noticing
An expired session shows up as failing AI features rather than as a login prompt. Open cursor.com, confirm you are signed in there, then sign out and back in inside the editor.
Cursor will not start at all
A blank window or a refusal to launch is not an outage, and checking the status page for it wastes time. The documented fixes:
| Symptom | Fix |
|---|---|
| Blank screen on launch | Quit fully and restart; then run Clear Editor History from the command palette |
| Blank screen persists on macOS | Move Cursor to the Trash and reinstall from cursor.com/download |
| Blank screen on Windows | Launch as administrator |
| macOS says the app is damaged | Quit, force-quit leftover processes in Activity Monitor, wait a minute, reopen |
| Damaged warning persists | Trash it, empty the Trash, reinstall; restart the Mac if needed |
| Sluggish and disk is nearly full | Remove unused extensions and run Clear Editor History |
| Stuck on an old build | Command palette, Cursor: Attempt Update, then restart |
Find out before you waste ten minutes
The worst version of an outage is the twenty minutes you spend assuming it is your code. Subscribe once and that stops happening.
Atom https://status.cursor.com/history.atom
RSS https://status.cursor.com/history.rss
Also email, SMS, Slack, Microsoft Teams, and webhooks- Slack or Teams is the highest-value option for a team: one person notices, everyone stops debugging.
- Webhooks if you want an incident to show up in your own dashboard or pause a pipeline.
- The feeds if you already read feeds. They carry the full incident history, which is also how you check whether a problem is recurring rather than new.
Questions people ask
Check status.cursor.com, which reports IDE, Cloud Agents, Automations, CLI, Bugbot, and cursor.com separately. No third-party page can answer this accurately, because they all cache. If Cursor is green, check the status page of the model provider you are using, since many Cursor incidents are upstream degradations.
Four things produce that message: the model provider is degraded, HTTP/2 is being blocked by a proxy or VPN, you are on a stale build, or the turn is genuinely large. Switch model first, then test on a phone hotspot, then switch HTTP Compatibility Mode to HTTP/1.1.
Six components: IDE, Cloud Agents, Automations, CLI, Bugbot, and cursor.com. Only one of them is usually affected in an incident, so identify which one covers the feature you are using before you decide to wait.
It is a crowd signal, not a source. It is useful for corroborating that other people are seeing the same thing at the same moment, particularly for regional or partial failures that never reach the official status page. Never treat it as the answer on its own.
Sometimes. The editor is local and keeps working when inference does not, so a dead chat with a healthy editor points at the IDE component or at the model provider. It can also be an exhausted credit pool, which looks similar and is not a fault.
Separate what is slow. Slow replies point upstream at the provider. Slow Tab or a sluggish editor points at local resources, extensions, or a very large workspace. Slow agent runs usually mean a large context or Max mode rather than anything being broken.
Set HTTP Compatibility Mode to HTTP/1.1 in Cursor Settings under Network, allowlist *.cursor.sh, *.cursor-cdn.com and *.cursorapi.com, and trust your proxy certificate if it does TLS inspection. Run Diagnostics in the same settings pane will tell you which of those is failing.
Subscribe on status.cursor.com by email, SMS, Slack, Microsoft Teams, or webhook, or follow the Atom feed at status.cursor.com/history.atom. For a team, Slack is the best value: one notification stops several people debugging the same non-bug.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.
- Cursor status components and incident history
- Cursor: network, proxy, and remote connections diagnostics, HTTP mode, allowlist
- Cursor troubleshooting guide launch and update failures
- Anthropic status