TL;DR

Open status.claude.com. If there's an active incident for Claude Code or the API, it's down for others too and the fix is to wait.

If the status page is green, read the exact error. Anthropic's error reference splits them into a few families:

Claude Code already retries transient failures up to 10 times with backoff before it shows you an error, so a visible error means the retries ran out.

The problem

Claude Code stops responding, or a red error appears mid-task. Searching "is Claude Code down" is the natural reflex, but a lot of the time the service is fine and something closer to home is wrong: a usage limit, a proxy, a VPN, an expired login. The two need opposite responses. Waiting doesn't fix a proxy, and reinstalling doesn't fix an outage.

Step 1: Check the Claude status page

status.claude.com lists incidents for Claude's services, including the API that Claude Code uses. Anthropic's error messages point there too. As of v2.1.198, when a 529 overload is the cause, the retry countdown names where to check status.

If you use Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or a gateway, the error names that provider or host instead. Check that provider's status page, not Anthropic's.

Step 2: Read the error message

API Error: 500 Internal server error

Anthropic describes it as a server-side issue, usually temporary. Try again in a moment. If it persists, check the status page.

API Error: Repeated 529 Overloaded errors

The API is at capacity. Anthropic's reference says a 529 isn't your usage limit and doesn't count against your quota. A message such as Opus is experiencing high load belongs to the same family. Switching to another model with /model can get you moving while the load passes.

Server is temporarily limiting requests (not your usage limit)

A short-lived throttle unrelated to your plan. Since v2.1.199 Claude Code retries it automatically before showing it. Wait briefly and try again.

You've hit your session limit or weekly limit

Not an outage. Your plan's allowance ran out, and the message names when it resets. See when Claude Code usage resets and what to do when you hit the limit.

Request rejected (429)

You hit the rate limit configured for your API key, Bedrock project or Google Cloud project. If a proxy or gateway answered with its own 429 page, its title appears after the ·.

Unable to connect to API

The connection to the API failed. Recent versions name the cause and keep the error code in parentheses, for example ConnectionRefused, ENOTFOUND or EHOSTUNREACH. Two of the messages:

Unable to connect to API. Check your internet connection
Couldn't connect through your proxy (ERR_PROXY_TUNNEL)
Connection dropped (ECONNRESET)

Anthropic lists no internet access, a VPN that blocks api.anthropic.com, and an unconfigured corporate proxy as common causes.

Step 3: Test the connection from the same shell

curl -I https://api.anthropic.com

On Windows PowerShell use curl.exe -I https://api.anthropic.com, so the built-in alias isn't used.

• curl fails: the problem is the network on this machine. Check Wi-Fi, VPN and DNS. Behind a corporate proxy, set HTTPS_PROXY before launching Claude Code.
• curl works but Claude Code doesn't: something between the runtime and the network. Check ANTHROPIC_BASE_URL if you use a gateway, and the firewall's allowed hosts in Anthropic's network configuration page.

Step 4: Rule out the local install

If the status page is green and the network is fine:

1. Update: claude update.
2. Check you're signed in to the account you expect: /status inside Claude Code.
3. Check for a stray ANTHROPIC_API_KEY in your environment, which overrides a subscription login.
4. For hangs or high CPU rather than errors, Anthropic's troubleshooting page covers the usual causes.

For unattended runs

In CI or other jobs nobody is watching, Anthropic documents CLAUDE_CODE_RETRY_WATCHDOG=1. It retries 429 and 529 capacity errors indefinitely instead of failing after the normal retries. A spend-limit 429 still fails at once. Pair it with a job timeout so a long outage doesn't hold a runner forever.

Telling a provider problem from a computer problem when you're away

When a session runs on a computer you're not sitting at, "Claude Code is down" can mean three different things: the provider is having trouble, the session hit a limit, or the computer itself went offline.

If you want to tell those apart from your phone, Happier is an open-source app that lets you run and orchestrate Claude Code, Codex and other agents cross-device, from your phone, desktop or the web.

What it shows:

• Each machine shows online or offline, with when it was last seen. If the computer is offline, the provider's status doesn't matter until it's back.
• Usage limits show as a banner in the session, Usage limit reached, with the reset time the provider gave. That separates a quota stop from an outage.

On the computer, happier daemon status shows whether the background service is running. Happier doesn't fix provider outages, network drops or a computer that's off. Setup is in Run Happier from your phone. It's on GitHub, you can download it here, and there's a Discord.

FAQ

Is Claude Code down right now?

Check status.claude.com. It's the source Anthropic's own error messages point to.

Does a 529 error use up my usage?

No. Anthropic's error reference says a 529 isn't your usage limit and doesn't count against your quota.

Claude chat works but Claude Code doesn't. Why?

They share an account but not always a path. Claude Code may be using an API key, a proxy, a gateway or a cloud provider. Check /status and your environment variables.

Why does Claude Code stop with "Connection lost while your computer was asleep"?

The computer slept during a request. Claude Code retries dropped connections, and if the turn still ends, send the message again.

Related guides

• How to tell which Claude Code, Codex and OpenCode session is stuck
• When does Claude Code usage reset?
• What to do when you hit the Claude Code usage limit
• How to keep Claude Code, Codex and OpenCode running when your laptop sleeps