Is Cursor Down? How to Check and What to Rule Out

Matching the symptom to the layer

Cursor failure symptoms and what to check first
What you see	Usual layer	First check
Agent requests hang or time out	Provider or Cursor incident	Check the status page first. If it is green, switch to a model from the other pool. A failure confined to one model points at that provider rather than at Cursor.
Tab suggestions stopped appearing	Local, usually	Tab runs against a different service from the agent. Reload the window, then confirm Tab is enabled for the current language before assuming an outage.
Cannot sign in, or signed out repeatedly	Auth or network	A corporate proxy or VPN blocking the auth callback produces exactly this. Try from a different network before filing anything.
A model disappeared from the picker	Stale build or plan	Model availability depends on your editor version and your plan. Update, then run agent models from the CLI to see what the account can actually call.
Everything is slow but nothing fails	Degraded performance	Status pages report degradation separately from outage, and it is easy to miss. Read the incident history, not just the headline colour.

Cursor is not one service

This is the thing that makes outages confusing. Agent requests, Tab completions, authentication, cloud agents and the dashboard are separate systems, and Cursor's documentation notes that models themselves are hosted by the model provider, a trusted partner, or Cursor, depending on which model you picked.

So "Cursor is down" is rarely the accurate description. More often one model provider is having a bad afternoon and every user who pinned that model is affected while everyone on Auto notices nothing.

The two-minute triage

Read the status page properly. Look at component health and recent incidents, not just the headline. Degraded performance is reported separately from an outage and is easy to scroll past.
Switch models. If a request fails on a frontier model and succeeds on Auto, you have found your answer and your workaround in one step.
Reload the window. A surprising share of "the agent stopped responding" is a stuck editor window rather than a stuck service.
Change networks. Corporate proxies and VPNs break authentication callbacks and long-lived connections in ways that look exactly like a server problem. A phone hotspot settles it in thirty seconds.
Check your version. A build several releases old can be missing a model or a feature entirely, and the symptom reads as a fault rather than as staleness.

If it is genuinely broken for you alone

Collect the request ID before anything else. In the CLI, /copy-request-id puts the last one on your clipboard, and /logs gives you the debug log path. Those two artefacts are the difference between a support conversation that resolves and one that consists of being asked to reproduce it.

The community forum is usually the fastest read during a live incident, because other people hit it at the same time and say so. For something reproducible and specific to you, the bug reporting guide explains what Cursor needs.

Links verified against cursor.com on 25 September 2026. Live status is only ever on status.cursor.com.

Frequently asked questions

Is Cursor down right now?

The authoritative answer is status.cursor.com, which reports component-level health and incident history. This page cannot tell you live status; it explains what to check and how to interpret what you find.

Where is the official Cursor status page?

status.cursor.com. It is linked from the footer of cursor.com and reports on individual components rather than a single site-wide indicator, which matters because the agent, Tab and authentication can fail independently.

The status page is green but Cursor is not working. What now?

Work down the layers. Switch models to see whether the failure follows a single provider. Reload the window. Check whether a VPN or corporate proxy is intercepting traffic. Then confirm your build is current, because a model or feature that launched recently will not work on an old client.

Why does only one model fail while the others work?

Because models are hosted by the model provider, a trusted partner, or Cursor, depending on the model. A single-model failure usually means an upstream provider incident rather than anything wrong with Cursor, and switching models is both the diagnosis and the workaround.

Does an outage consume my usage allowance?

A failed request that returns nothing is not the same as a completed one. If you see usage that does not match the work you got out of it, the usage dashboard and a request ID are what support will ask for. Copy the latter with /copy-request-id in the CLI.

Where do I report a Cursor outage?

The status page is for reading, not reporting. Cursor's forum is where users compare notes quickly during an incident, and the bug reporting process in the help centre is the route for something reproducible that is specific to you.