Official Status Page
Anthropic's status page lives at status.claude.com (also accessible at status.anthropic.com). It tracks three components independently:
Each component has its own status indicator. Claude Code can be down while the API is operational, or vice versa. The API powers Claude Code under the hood, so API outages always affect Claude Code, but Claude Code can have its own issues (login failures, CLI bugs) independent of the API.
Subscribe to Status Updates
- Email/SMS: Get notified when incidents are created, updated, or resolved
- RSS Feed: Useful for automated monitoring and custom dashboards
- Webhook: Route alerts to Slack, PagerDuty, or any HTTP endpoint
- Third-party trackers: StatusGator, IncidentHub, and IsDown aggregate Anthropic data with historical uptime logs
Common Claude Code Issues
Not every problem is an outage. Most Claude Code disruptions fall into one of five categories, and only one of them means Anthropic's servers are actually down.
| Issue | Error Code | What It Means |
|---|---|---|
| Rate limit exceeded | 429 | You hit your plan's request cap. Claude Code is working fine. |
| Authentication failure | 401 / 403 | API key is invalid, expired, or missing from your environment |
| Network timeout | Timeout / ECONNRESET | Request took too long. Could be your network, DNS, or server load |
| Model degradation | 200 (slow) | Responses arrive but with higher latency or lower quality |
| Infrastructure outage | 500 / 503 | Anthropic's servers are down. Confirmed on the status page |
The first three are local problems you can fix yourself. Model degradation is harder to detect because you get valid responses, just slower or worse ones. Infrastructure outages are the only category where you genuinely need to wait for Anthropic.
Rate Limits vs Outages
The 429 Too Many Requests error is the single most common reason developers think Claude Code is "down." It isn't. A 429 means you've exceeded your plan's request allowance. Anthropic uses rate limits to ensure fair access across all users.
Rate Limit (429)
Claude Code is working. You're sending requests faster than your tier allows. The response headers include a retry-after value telling you when the limit resets. Wait, reduce frequency, or upgrade your plan.
Outage (500/503)
Claude Code is actually down. Anthropic's infrastructure has a problem. Check status.claude.com for details. No amount of retrying will fix it. Switch to a fallback workflow until the incident is resolved.
How to Read Rate Limit Headers
When you get a 429 response, check these headers:
- retry-after: Seconds until your limit resets
- x-ratelimit-limit: Your plan's maximum requests per minute
- x-ratelimit-remaining: Requests left in the current window
- x-ratelimit-reset: Timestamp when the window resets
Free tier users hit rate limits regularly during extended coding sessions. Pro plan users ($20/month for Claude Pro, or pay-per-token on the API) have significantly higher limits but can still trigger them with rapid-fire requests. If you hit limits consistently, the problem isn't Claude Code. It's your usage pattern.
Recent Incidents (March 2026)
Claude Code has had several notable incidents in early 2026. Understanding what happened helps set expectations for future disruptions.
| Date | Issue | Duration / Resolution |
|---|---|---|
| March 11, 2026 | Elevated errors on Claude.ai and Claude Code login failures | ~3 hours. Resolved by 17:56 UTC |
| March 8, 2026 | Filesystem connector missing from Claude Desktop | Restored to allowlists same day |
| March 7, 2026 | Elevated errors on Claude Haiku 4.5 | ~1 hour (13:35 - 14:44 UTC) |
| March 2, 2026 | Widespread outage affecting Claude.ai, auth, and API endpoints | Several hours. Full resolution same day |
| Late February | Usage reporting not returning data after Feb 25 18:40 UTC | Fix deployed over several days |
The March 2 outage was the most severe, taking down web access, authentication, and specific model endpoints simultaneously. The March 11 incident was narrower, primarily affecting Claude Code logins while the API remained functional. Both were resolved within hours.
Diagnostic Checklist
When Claude Code stops responding, work through these steps in order. Each eliminates a category of problem.
1. Check status.claude.com
If it shows degraded or outage, the problem is on Anthropic's end. Nothing you can do except wait or switch to a fallback. If it shows operational, the problem is on your side.
2. Test Your API Key
Run a direct curl request to the Claude API. If it returns a valid response, the API is fine and the issue is in Claude Code's CLI layer specifically.
3. Update Claude Code
Run npm install -g @anthropic-ai/claude-code to get the latest version. CLI bugs are fixed frequently, and running an outdated version can cause errors that look like outages.
4. Check Rate Limits
Look for 429 responses and check the retry-after header. If you're rate-limited, reduce request frequency or wait for the window to reset.
5. Test Your Network
Run a traceroute to api.anthropic.com. If packets drop before reaching Anthropic, the problem is your ISP or local network configuration, not Claude Code.
6. Check Environment Variables
Verify ANTHROPIC_API_KEY is set correctly in your shell. A missing or malformed key causes 401/403 errors that can look like an outage.
Quick Test Command
Test your API key directly (replace YOUR_KEY with your actual key):
curl https://api.anthropic.com/v1/messages -H "x-api-key: $ANTHROPIC_API_KEY" -H "anthropic-version: 2023-06-01" -H "content-type: application/json" -d '{"model":"claude-sonnet-4-20250514","max_tokens":10,"messages":[{"role":"user","content":"hi"}]}'
Fallback Workflows When Claude Code Is Down
Claude Code depends entirely on Anthropic's infrastructure. When that infrastructure has problems, everything stops. The fix isn't to wait. It's to have parts of your workflow that run on independent infrastructure.
Two operations dominate AI coding agent time: searching the codebase for context (roughly 60% of agent runtime, per Cognition's measurements) and applying code edits. Both can run on separate infrastructure.
Morph Fast Apply
Fast Apply is a purpose-built model for code edits. It runs at 10,500+ tokens/sec on Morph's own GPU cluster, not through Anthropic. When Claude Code is down, Fast Apply keeps working because it has zero dependency on Anthropic's servers. It connects via standard API to Cursor, Windsurf, Cline, or any tool that supports custom model endpoints.
WarpGrep
WarpGrep is a semantic code search engine. It indexes your codebase and serves results through an MCP server. The indexing and search run on Morph's infrastructure, independent of Anthropic. When Claude Code can't search your codebase, WarpGrep still can. It works with any MCP-compatible agent.
Why Independent Infrastructure Matters
Single-provider dependency is the root cause of downtime-induced productivity loss. If your code search, code editing, and agent orchestration all run through one provider, a single outage blocks everything. Distributing these operations across providers with independent infrastructure means any one provider's outage only degrades part of your workflow, not all of it.
Reducing Downtime Risk
Beyond having fallbacks, you can reduce how often Claude Code disruptions hit you in the first place.
Offload Search to WarpGrep
Claude Code spends 60% of its time on search operations. Each search is an API call to Anthropic. Running search through WarpGrep's MCP server means fewer Anthropic API calls, which means fewer rate limit hits and less exposure to API outages.
Use Fast Apply for Edits
Code application is the second biggest consumer of API calls. Running edits through Fast Apply at 10,500 tok/s offloads that traffic from Anthropic's servers. Your agent still orchestrates, but the heavy compute happens elsewhere.
Set Up Status Alerts
Subscribe to status.claude.com webhooks. Route them to Slack or PagerDuty. Knowing about an outage 5 minutes earlier means you can switch workflows before losing momentum.
Keep Claude Code Updated
Many 'outage' reports are actually CLI bugs fixed in newer versions. Run npm install -g @anthropic-ai/claude-code regularly. Check the changelog at the Anthropic release notes page before reporting issues.
The pattern is straightforward: the fewer API calls you send to any single provider, the less vulnerable you are to that provider's downtime. Specialized subagents for search and code application reduce your Anthropic API surface area while also being faster at their specific tasks.
Frequently Asked Questions
Where do I check Claude Code status?
The official status page is status.claude.com (also accessible at status.anthropic.com). It tracks Claude.ai, the Claude API, and Claude Code as separate components. Subscribe via email, SMS, RSS, or webhook for notifications.
Is a 429 error the same as Claude Code being down?
No. A 429 means you hit your plan's rate limit. Claude Code is working fine. You're sending requests faster than your tier allows. Check the retry-after header to see when the limit resets. Reduce request frequency or upgrade your plan.
How often does Claude Code go down?
Major outages are infrequent but not rare. March 2026 saw two notable incidents: a widespread outage on March 2 lasting several hours, and login issues on March 11 lasting about 3 hours. Model-specific degradation (slower responses, elevated error rates) happens more frequently but is harder to detect.
What should I do when Claude Code is down?
Confirm the outage at status.claude.com. If it's an infrastructure issue, switch to tools on independent infrastructure. Morph Fast Apply handles code edits at 10,500 tok/s on its own GPU cluster. WarpGrep provides semantic code search via MCP. Neither depends on Anthropic.
Can I get notified when Claude Code goes down?
Yes. Subscribe at status.claude.com via email, SMS, RSS, or webhook. Webhooks route to Slack, PagerDuty, or any HTTP endpoint. Third-party services like StatusGator and IncidentHub provide additional historical uptime data.
Related Guides
Keep Working When Claude Code Is Down
Morph Fast Apply and WarpGrep run on independent infrastructure. No Anthropic dependency. 10,500 tok/s code edits and semantic search that work regardless of Claude Code status.