Diagnostic entry point · 40 error strings indexed
Claude Code isn't working. Start here.
Most Claude Code failures are environmental, not model bugs — and the fastest route to the cause is not a new tool. It is running the checks that already ship, then matching the exact string your terminal printed against what 12,045 other reports say about it.
40 error strings · 12,045 reports indexed · 2,443 workaround leads · data through August 30, 2026
Four steps, in this order
Ordered by how much each step rules out. Step 1 separates service failures from account limits and malformed requests before any local configuration is changed.
- Classify the API response before touching your config
A 500 or 529 is service-scoped: check service status and wait before changing local files. A 429 is account-scoped: stop parallel retries and check usage, plan limits and shared credentials. A 400 is api-scoped: inspect the malformed request instead of waiting for an outage to clear.
Check current issue-report activity → - Run the built-in checker
Claude Code ships its own diagnostic: /doctor in session (/checkup since v2.1.205), or claude doctor from the terminal when a session will not start. It validates installation health, settings files and hook schemas.
- Look up your exact error string
Copy the error line verbatim and find it among 40 indexed strings. Each page shows how many people report it, whether it is growing or converging, which builds are affected, and the workarounds posted in the threads.
Browse errors by exact message → - Narrow by platform when the error is environment-specific
Path, permission and MCP failures often depend on the OS and install method. 95 platform-by-area pages separate a Windows-only fault from a universal one.
Browse platform-specific reports →
The checks that already ship
Claude Code has a built-in diagnostic. Run these before installing anything — they cost nothing and cover the common faults. Each row also states what the command does not tell you, because a clean report is routinely mistaken for proof that configuration is fine.
-
/doctoralias/checkupin sessionChecks: Installation health (including duplicate installs), invalid settings files, hook schema errors, unused extensions, duplicate subagent names in one directory, and checked-in CLAUDE.md content Claude could infer from the codebase.
Does not: Does not resolve which settings scope won, does not flag a value silently overridden by a closer scope or an environment variable, and does not trace why a registered hook fails to fire.
-
claude doctorterminalChecks: The same installation and settings diagnostics, read-only, without starting a session.
Does not: Proposes nothing and changes nothing — use it when a session will not start at all.
-
/statusin sessionChecks: Which settings sources are active, and whether managed (enterprise) settings apply.
Does not: Does not validate the contents of those sources.
-
/mcpin sessionChecks: Live MCP server connection state and pending approvals.
Does not: Does not test the server command outside Claude Code — run it yourself to see its stderr.
-
/permissionsin sessionChecks: The resolved allow/deny rules actually in force, after all scopes merge.
Does not: Does not explain which file contributed a rule.
-
/contextin sessionChecks: What really loaded into the context window. /memory and /skills break it down further.
Does not: Does not tell you why something was dropped when the budget ran out.
-
/hooksin sessionChecks: Which hooks are registered.
Does not: Does not show why a registered hook never fires — use `claude --debug hooks` for that.
Command coverage per Anthropic's
troubleshooting documentation.
Commands and aliases change between releases — /help in your build is authoritative.
Find your error, then read its scope
Paste your error line into search, or pick the scope it belongs to below. The scope is the most useful thing to learn first: it tells you whether to wait, re-authenticate, check your network, or edit local configuration.
-
Anthropic service
-
API Error: 500603 reports · 11 open -
API Error: 529 overloaded_error254 reports · 27 open
-
-
API request
-
API Error: 400 invalid_request_error2,591 reports · 129 open -
not_found_error: model83 reports · 4 open
-
-
Account, auth & limits
-
API Error: 429690 reports · 23 open -
API Error: 401441 reports · 48 open -
compaction failed379 reports · 31 open -
OAuth token has expired344 reports · 91 open
+5 more account, auth & limits strings · see all 40 →
-
-
Session & context
-
request_too_large38 reports · 2 open
-
-
Network, proxy & TLS
-
ECONNRESET: socket hang up229 reports · 70 open -
ECONNREFUSED207 reports · 9 open -
getaddrinfo ENOTFOUND49 reports · 17 open -
self-signed certificate in certificate chain41 reports · 6 open
+2 more network, proxy & tls strings · see all 40 →
-
-
Tooling & MCP
-
garbled terminal output829 reports · 148 open -
spawn ENOENT / command not found506 reports · 58 open -
MCP server failed to connect229 reports · 45 open
-
-
Local machine
-
AbortError: The operation was aborted2,101 reports · 47 open -
ENOENT: no such file or directory1,200 reports · 64 open -
SyntaxError: Unexpected token698 reports · 30 open -
EISDIR: illegal operation on a directory495 reports · 9 open
+13 more local machine strings · see all 40 →
-
No clean error line? Start from the surface
Silent failures — a hook that never fires, an MCP server that never appears — often print nothing. These are the areas where configuration problems actually surface, ranked by how many reports are still open.
When the built-in checker says you're fine
138 issues in this tracker name /doctor,
/checkup or claude doctor in their title —
6 still open. False positives on valid config, misreported install
paths and checks that hang are all documented. A clean report narrows the search; it does not
end it.
- [BUG] ✗ Auto-update failed · Try claude doctor or npm i -g @anthropic-ai/claude-code
- [BUG] 1.1617.0 (8d6345): New Cowork sessions crash with "exited with code 1" - /doctor and /bug non-functional
- [BUG] claude doctor hangs indefinitely, uses 100% CPU, and soft locks terminal session
- claude doctor command freezes with no response to Enter key
- [BUG] claude doctor reports insufficient permissions for automatic updates when using NVM
- [BUG] Standalone CLI binary missing seccomp binaries — /doctor reports "seccomp not available"
Ranked by community engagement, from the 34 reports with a full page here.
Common questions
- Is there a one-command fix for Claude Code?
- Yes, and it ships with Claude Code: run /doctor in session (aliased /checkup since v2.1.205) or claude doctor from the terminal when a session will not start. It validates installation health, settings files, hook schemas and unused extensions. What it does not do is resolve which settings scope won, flag a value overridden by an environment variable, or trace why a registered hook fails to fire — those still need the manual steps on this page.
- What should I do when /doctor reports everything is fine but Claude Code is still broken?
- Move to the error string. Copy the exact line your terminal printed and look it up among 40 indexed strings covering 12,045 reports. The triage scope on each page tells you whether the cause is the service, your account, the network, an integration or your machine — which decides whether local edits can help at all.
- Does /doctor ever get it wrong?
- It can. 138 issues in this tracker name /doctor, /checkup or claude doctor in their title, 6 of them still open — including false positives on valid MCP config, misreported install paths and keychain misdiagnosis. Treat a clean report as one data point, not proof.
- Do you upload my configuration anywhere?
- No. This page is static HTML with no analytics and no telemetry. There is nothing to install and nothing runs against your machine — the checks listed here are Anthropic's own commands, executed locally by you.
- Is this an official Anthropic troubleshooting page?
- No. This is an independent community project, not affiliated with or endorsed by Anthropic. The built-in command descriptions come from Anthropic's published documentation; everything else is derived from the public anthropics/claude-code issue tracker.