Get your free AI audit.
Book a call →Anthropic runs six services. Here is the live endpoint to check, the four sources that confirm it, and the five-step routine to keep working.
Before you debug Claude Code, check whether Claude itself is having an incident. Anthropic's public status page is the first check. If it is operational, move through the local CLI, account, network and API-key checks below. This draft is a refresh of the existing canonical page, so every live status and incident claim must be rechecked on the day of publication.
This is the long-form companion to the one-screen Claude status page on the site. The status page is the live badge. This post is the playbook. If you just want the green or red light, the badge is faster. If you want to know what the badge actually means, what to do when it is red, and which alternative to switch to in the meantime, keep reading.
Check Anthropic's live status page first. It is the authoritative source for active incidents and maintenance. The result changes, so this article does not hardcode a current green or red status.
If you are reading this and Claude is sluggish, the badge is your first stop. The full status payload is at https://status.anthropic.com/api/v2/summary.json. Pull it from any terminal:
curl -sL https://status.anthropic.com/api/v2/summary.json | jq '.status'
That returns an indicator such as none, minor, major, or critical. A non-none indicator means the status page is reporting degradation. Use the component details and incident timeline for scope rather than treating a timestamp as a freshness guarantee.
A real operator will not stop at the green badge. The next question is why is Claude not responding. The status page tells you if Anthropic is up. It does not tell you if your account is rate limited, your network is on a flaky route, or your CLI rolled back to a broken release. Those are local problems the status page cannot see.
Use a simple source hierarchy rather than a majority vote.
1. Anthropic's official status page at https://status.anthropic.com. Component-level breakdown, scheduled maintenance, and a 90-day incident history. 2. The two public JSON endpoints that back the status page. api/v2/summary.json is the current state. api/v2/incidents.json is the 50 most-recent incidents with full timelines. Both are unauthenticated. Both are the same data the page renders. 3. Your own reproducible request or login test, with the error body preserved. 4. Community sources such as Reddit, Discord and X, used as context only.
Save the curl command above to a shell alias. When Claude hangs, run the alias. If the JSON does not load, check your connection. If the JSON loads and shows none, move on to account, CLI, network and request-level checks.
For the broader picture when the API is yellow, the Claude status page on the site polls the same endpoint every 60 seconds and lists the last five incidents underneath. The short checklist version is the step-by-step triage I wrote last month. Pair the curl above with the checklist for the four-line routine.
The likely causes are:
429 can indicate a rate limit, usage limit or acceleration limit. Check the error body and current plan or API documentation rather than relying on a fixed message count.claude --version, then check the current Claude Code setup and troubleshooting documentation before updating.If three of these are ruled out and the status page is green, the issue is something subtler. Reinstall Claude Code, clear your browser cache, or move to a different network. The diagnostic value of the status page is that it lets you skip these last two steps when Anthropic is, in fact, down.
The Claude API and Claude Code can fail independently. Claude Code can use Anthropic's API and other supported provider paths, while claude.ai and the Console have their own product surfaces. A problem in one path does not automatically take every Claude surface down.
Claude Code can break on its own while the API is fine. The CLI is a separate binary with its own release cadence. A bad release shows up as a Claude Code component incident on the status page (look for the row labelled "Claude Code", not just "Claude API"). Last month the Claude Code component was operational during a Sonnet-3.5 deprecation incident that broke every Sonnet-3.5 call through the API. The Claude API status was yellow. Claude Code was green. Same calendar, different services.
How to tell which one you are looking at:
claude --version and the release notes. If those are clean, the Claude Code component on the status page is your source of truth.The status page components list is the fastest way to disambiguate. Read the six components, not just the top banner.
Anthropic's incident timeline shows the current scope and latest updates. There is no reliable universal duration, so use the incident updates rather than a fixed waiting rule.
Read the incident description and updates instead of trying to predict the resolution time from old incidents.
For work that has to ship today, the answer is not to wait. Switch. The five working alternatives when Claude is down are listed below. Pick one and keep moving.
The status feed is the official source. It is also the only thing Anthropic controls. Two things it does not cover:
incident_updates array in the JSON gives you a clearer timeline than the rendered page.For Claude Code specifically, the limitation is the release-skew problem. The status page will say all green because the API and the web app are fine, while the CLI you updated last Tuesday is broken on a new release. The Claude Code component on the status page covers major incidents. It does not catch every bad release.
Five fallbacks, ranked by how much work they take to switch.
1. Wait if the work is not urgent and the incident updates show active recovery. 2. Switch to a different Claude model. If Opus 5 is degraded but Sonnet 4 is fine, change your prompt or your release notes and keep going. The model picker is the cheapest switch. 3. Switch to a different Claude surface. If the API is degraded but Claude Pro is fine, log in on the web app and use the chat interface. If the web app is degraded but the API is fine, route your work through the API with a CLI or a script. 4. Use another supported provider path only if your deployment and the provider's current documentation support it. Do not treat ANTHROPIC_API_KEY as a generic provider switch. 5. Use a local model only if you already have a tested local setup. It may be slower and its output quality will vary.
The rule I use is simpler: check the official incident updates, then choose between waiting, a supported alternate path, or a tested local fallback based on how urgent the work is.
Yes, if you build with Claude Code daily. The five-second curl that prefixes this post is the cheap win. The Anthropic status page is the result. The two JSON endpoints are the data. The four sources are the corroboration. The five alternatives are the fallbacks. There is no good reason to install a third-party "Claude down detector" when the official status page is one curl away.
If you only use Claude occasionally, the Anthropic status page is the answer. Bookmark it. Check it when Claude hangs. If it is green, look at your account, your CLI version, or your network. If it is yellow, switch to one of the fallbacks for the next two hours and check back.
The conversational "is Claude down right now" question has two answers, and the answer that's right for you is the one that gets you back to shipping.
If you found this useful, the step-by-step checklist for figuring out whether Claude is down is the next read. It is the 4-minute triage you should run before diving into the curl commands above. For the live status badge, the Claude status page polls the same API every 60 seconds and shows the last five incidents underneath.
If your Claude Code session is breaking on something other than the API (release-skew bugs, model deprecation, MCP errors), the Claude Code plan mode checklist is the closest thing to a "is this a me problem or a Claude problem" diagnostic. For the broader question of how to keep shipping when the API is fluctuating, the Claude Code tips from 1,000 hours in covers the patterns I lean on when the tooling is rough.
Claude is up if the badge at https://status.anthropic.com is green. The page tracks six components (claude.ai, Claude Console, Claude API, Claude Code, Claude Cowork, Claude for Government). If any component is yellow, orange, or red, that surface is degraded. The full JSON payload is at https://status.anthropic.com/api/v2/summary.json. If you see All Systems Operational and Claude is still not responding, the problem is on your side: account rate limit, local CLI version, browser session, or network routing.
The fastest way is curl -sL https://status.anthropic.com/api/v2/summary.json | jq '.status'. The result includes an indicator such as none, minor, major or critical. If you cannot reach the endpoint, check your network. If the endpoint is reachable and shows none, move on to account, CLI, browser and request-level checks.
Check the "Claude Code" component row on https://status.anthropic.com alongside the API row. If the row is operational and Claude Code is still broken, run claude --version and check the current setup and troubleshooting documentation.
Claude's current status is represented by the indicator in the JSON payload at https://status.anthropic.com/api/v2/summary.json. A none indicator means the page is reporting no active degradation. For any other indicator, read the component details and incident updates on the rendered status page.
Common causes include account limits, local CLI version skew, an expired browser session, a provider or network-path issue, and an Anthropic incident. If the status page reports no degradation, check the local and account-level causes rather than assuming the service is down.
Check the "Claude API (api.anthropic.com)" component row on the status page. A real authenticated request also tests your credentials, network, request format and provider path, so do not treat an unauthenticated HEAD request as a complete health check.
There is no reliable universal duration. Use Anthropic's incident updates for the current scope and recovery status instead of planning around a fixed waiting time.
The cheap one is the Anthropic status page itself. The Python one is a 10-line script that polls https://status.anthropic.com/api/v2/summary.json every 60 seconds and posts to a Slack channel when the indicator is non-zero. The hosted ones (StatusGator, Atlassian Statuspage readers) are fine but rarely necessary. The official status page is one curl away and has the most accurate data.
If the badge is green at https://status.anthropic.com, the page is reporting no active degradation at that moment. Check the component details and incident timeline for scope. The incidents.json endpoint has historical incident records, not a substitute for the live summary.
Claude Code is fine to use during a partial outage as long as the Claude Code component on the status page is green. The CLI is independent of the API for non-API components (local skill execution, MCP server calls, file operations). If the API is yellow but Claude Code is green, switch models inside the CLI and keep working. If Claude Code is yellow, do not push updates. Wait for the component to flip back.
Claude going down kills momentum. The hourly-rate cost of an unscheduled two-hour outage is real for anyone shipping daily. If Claude is the centerpiece of your workflow, the Claude Code Blueprint is the 60-minute setup that turns the API from a single point of failure into a system you can keep running. Five lessons. Free. No coding required.
Five interactive lessons. Install Claude Code, build your first automation, and deploy it live on the internet — all in under an hour. Free, no coding required.
Grab the Blueprint →