Claude Code down? How to check, and keep working
Claude Code down? Check status.claude.com, /status and /doctor in 30 seconds, tell an outage from your own limit or network, and keep working meanwhile.
Contents
When Claude Code seems down, open status.claude.com first. Claude Code, the Claude API and claude.ai each get a row there, with any open incident pinned above them. An incident on the Claude Code row means it really is down: wait, or work elsewhere for a while. With every row green, type /status and /doctor and read the exact error, because then the usual cause is your usage limit or your network.
The commands, error text and quotes on this page are from Anthropic's docs as they read on 22 October 2026. The whole check takes about 30 seconds.
Check in 30 seconds#
- Open status.claude.com and look at the Claude Code and Claude API rows. Subscribe there and the next incident reaches you by email, SMS, Slack, Microsoft Teams, webhook or RSS.
- Type
/statusin Claude Code. It prints "version, model, account, and connectivity", and it answers even while Claude is still responding (commands). - Type
/doctor. It checks the installation, settings, extensions and context use, and proposes fixes. Whenclaudewon't start at all,claude doctorin your shell does the same (troubleshooting). - Read the error word for word, because the wording says whose problem it is.
Outage, your limit, or your network?#
| The message says | What it means | What to do |
|---|---|---|
API Error: 500 Internal server error (or another 5xx) |
A failure inside Anthropic's API, "not caused by your prompt, settings, or account" | Check the status page, wait a minute, then type try again |
API Error: Repeated 529 Overloaded errors |
The API is at capacity. "A 529 is not your usage limit and doesn't count against your quota" | Wait a few minutes, or /model to another model: capacity is tracked per model. More in Claude Code 529 overloaded |
You've hit your session limit · resets 3:45pm (or weekly) |
Your plan's limit, not an outage | Wait for the reset, use usage credits, or another account. See Claude limit reached: what to do |
You've hit your Opus limit |
A limit for one model family | /model to a model outside that family |
Unable to connect to API, ENOTFOUND, ECONNRESET |
Your network, VPN, proxy or firewall | curl -I https://api.anthropic.com from the same shell; check HTTPS_PROXY and any leftover ANTHROPIC_BASE_URL. More in Claude Code unable to connect to API |
Login expired · Please run /login |
Your sign-in, not the service | /login again |
Every row is from Anthropic's error reference. By the time a 5xx or 529 reaches your screen, Claude Code has already retried it "up to 10 times with exponential backoff". And Connection refused while curl gets through usually means ANTHROPIC_BASE_URL still points at a proxy that is no longer running.
If you suspect your limit rather than an outage, /usage shows the plan's usage bars and when each one resets.
What to do while it's down#
Wait it out, without babysitting#
Your session is saved on your computer. When the service is back, claude --resume in the same folder opens it again, with your last message still in the conversation.
For runs nobody is watching, Anthropic documents two environment variables:
CLAUDE_CODE_RETRY_WATCHDOG=1retries429and529capacity errors "indefinitely instead of failing". Anthropic suggests it for CI jobs.API_TIMEOUT_MSis the per-request timeout in milliseconds, 600000 (10 minutes) unless you raise it for a slow network or proxy.
Switch model#
Anthropic tracks capacity per model, so a 529 on Opus says nothing about Sonnet. /model switches, and you carry on in the same conversation. The first request on the new model reads the whole conversation again with no cache hits, so that one request costs a little more.
Use another provider's agent#
Codex runs on OpenAI's service with its own limits, so a ChatGPT subscription keeps working through a Claude incident. The two agents work differently: Claude Code vs Codex says where each is stronger.
Run a local model#
Claude Code talks to any Anthropic-compatible address, and Ollama documents two ways in (Ollama docs). ollama launch claude does it in one command. By hand, you set ANTHROPIC_BASE_URL=http://localhost:11434, ANTHROPIC_AUTH_TOKEN=ollama and an empty ANTHROPIC_API_KEY, then start claude --model qwen3.5. The model runs on your own hardware, so a Claude outage and a dead internet connection both leave it running. It is also far weaker than Claude on hard work, and Ollama asks for a context window of 64k tokens or more. Unset the three variables when Claude is back. The step-by-step setup is in Claude Code with Ollama.
How Shuriken handles it#
Shuriken runs Claude Code on your own computers, so an Anthropic outage reaches it too. Two things differ around the outage.
- When the network drops or a model is refused, the chat waits and continues, or moves on, instead of dying mid-task. Since 3.3, a refused account moves the chat to another of your sources (the story).
- A Codex or Gemini CLI chat in the same project runs on its own accounts, so you can keep working there (Claude Code and Codex in one project). Shuriken never turns a Claude Code chat into a Codex chat. Moving work between agents is your call.
Try the free tools: Claude limit reset tracker · Codex limit reset tracker
Sources
- Claude Status (accessed 22 October 2026)
- Error reference (Claude Code docs) (accessed 22 October 2026)
- Troubleshooting (Claude Code docs) (accessed 22 October 2026)
- Commands (Claude Code docs) (accessed 22 October 2026)
- Claude Code with Ollama (Ollama docs) (accessed 22 October 2026)
Frequently asked questions
How do I know if Claude Code is down?
status.claude.com answers it: claude.ai, the Claude API and Claude Code are separate components there, with any incident at the top. When all three are green, look at your limit, your network or your setup instead; /status in Claude Code shows the account and connectivity.
Is a 529 overloaded error my usage limit?
No. Anthropic's docs say a 529 is not your usage limit and doesn't count against your quota: the API is at capacity, and Claude Code has already retried several times before API Error: Repeated 529 Overloaded errors reaches you. Wait a few minutes, or /model to another model, since capacity is tracked per model.
Does Claude Code retry by itself when the API fails?
Yes, up to 10 times with exponential backoff for a temporary failure, and the spinner shows Retrying in Ns · attempt x/y while it does. CLAUDE_CODE_MAX_RETRIES changes the count, and CLAUDE_CODE_RETRY_WATCHDOG=1 makes unattended runs keep retrying 429 and 529 errors.
Can I use Claude Code with another model while Claude is down?
Yes, by pointing Claude Code at another Anthropic-compatible endpoint, such as a model running in Ollama on your own computer. Expect a weaker model than Claude and set a context window of at least 64k tokens, as Ollama advises.
Will I lose my conversation if Claude Code goes down mid-task?
No. The session is stored on your computer, not on Anthropic's side of the outage. Once the service is back, claude --resume in the same folder picks it up with your last message still in it, and typing try again sends the work on.
