Claude Code ECONNRESET Fix: Every Cause and Solution (2026 Complete Guide)

You're mid-task in Claude Code. Everything was working fine. Then the terminal spits out:

API Error: Unable to connect to API (ECONNRESET)

You hit retry. Same error. You restart Claude Code. Same error. You check status.anthropic.com and it shows everything green. So what is actually happening?

ECONNRESET is one of the most frustrating Claude Code errors because it looks like an infrastructure problem Anthropic's servers must be down, right? but in the majority of cases, it's something on your end: a VPN intercepting TLS, an expired auth token, a proxy blocking the connection, or your network taking a broken path to Anthropic's API endpoints. The official status page showing green is actually a clue that the problem is local, not global.

This guide covers every real cause I've seen and fixed, with exact commands and steps for each one. Start from the top and work down most people hit the fix within the first two or three sections.

• What ECONNRESET actually means in Claude Code's context

• The 30-second diagnostic to tell local vs infrastructure problem apart

• Fix for VPN and corporate proxy conflicts the #1 cause in 2026

• Fix for expired or corrupted auth tokens

• Fix for firewall and network path issues

• Fix for outdated Claude Code version conflicts

• What to do when none of the above work

What ECONNRESET Actually Means

ECONNRESET is a Node.js error code it stands for "Error: Connection Reset." It means a TCP connection was established between your machine and Anthropic's API server, but the server sent a RST (reset) packet and closed the connection before finishing the response. It's different from a timeout (where no connection was established) and different from a 500 error (where the server responded but with an error).

The RST packet usually happens for one of three reasons:

• The server is overloaded and shedding connections it can't handle infrastructure-side issue

• Something in the network path (VPN, proxy, firewall) is intercepting and terminating the TLS handshake before it completes

• The auth token in the request is malformed or expired, causing the server to reject the connection at the transport layer

When Anthropic's status page shows green, you can rule out case one. That leaves cases two and three both of which are fixable on your end in minutes. Let's diagnose which one you're dealing with.

Step 1 : The 30-Second Diagnostic

Before trying any fix, run this curl command from the same machine and same network where Claude Code is failing:

curl -v 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-haiku-4-5-20251001","max_tokens":10,"messages":[{"role":"user","content":"hi"}]}'

Three possible results each points to a different cause:

• You get a JSON response (even a 400 or 401 error): The API is reachable. The problem is in Claude Code's auth token or configuration, not your network. Jump to the auth token fix section.

• You get a TLS handshake error or "curl: (35) OpenSSL SSL_connect": Something in your network is intercepting HTTPS traffic almost certainly a VPN or proxy. Jump to the VPN fix section.

• You get "curl: (7) Failed to connect" or a connection timeout: Your network can't reach Anthropic's servers at all. Jump to the firewall and network path section.

This one diagnostic tells you exactly which section to read. Don't skip it it saves 20 minutes of trying fixes that won't work for your specific situation.

Fix: VPN and Corporate Proxy Conflicts (Most Common Cause)

This is the #1 cause of Claude Code ECONNRESET in 2026, especially for developers working from corporate networks, co-working spaces, or with personal VPNs enabled. Here's why it happens and how to fix it.

Many VPNs and enterprise proxies perform TLS inspection they intercept encrypted HTTPS connections, decrypt them, inspect the traffic, and re-encrypt before forwarding. This is done for security monitoring purposes. The problem: Anthropic's API uses certificate pinning and strict TLS verification. When a proxy substitutes its own certificate, Claude Code's TLS verification fails, the connection gets reset, and you see ECONNRESET.

Quick Test First

Disable your VPN completely and try Claude Code again. If it works without VPN, you've confirmed the cause. The fix is not "stop using VPN" it's adding Anthropic's API domain to your VPN's bypass/exclusion list.

Personal VPN (ExpressVPN, NordVPN, Mullvad, etc.)

Most personal VPNs have a "split tunneling" or "bypass list" feature. Add these two domains to the bypass list so they route directly without VPN inspection:

api.anthropic.com
claude.ai

Where to find this setting: VPN app → Settings → Split Tunneling (or "Bypass" / "Exclusions"). The exact label varies by provider but every major VPN supports this. After adding the exclusions, reconnect VPN and test Claude Code.

Corporate Network / Enterprise Proxy

This requires your IT team to whitelist Anthropic's domains on the proxy. Send them this exact list:

api.anthropic.com        (HTTPS, port 443)
claude.ai                (HTTPS, port 443)
anthropic.com            (HTTPS, port 443)

While waiting for IT, test on mobile hotspot this routes around the corporate proxy entirely. If Claude Code works on hotspot, you've confirmed the corporate proxy is the issue and have a working workaround until IT resolves it.

You can also configure Claude Code to use a specific proxy explicitly, which sometimes helps even on corporate networks when the auto-detection is wrong:

# Set proxy for Claude Code
export HTTPS_PROXY=http://your-corporate-proxy:port
export HTTP_PROXY=http://your-corporate-proxy:port

# Or add to your shell profile permanently
echo 'export HTTPS_PROXY=http://your-proxy:port' >> ~/.zshrc

Fix: Expired or Corrupted Auth Token

Claude Code stores your Anthropic authentication token locally. If this token expires, gets corrupted during an update, or was set incorrectly during initial setup, the API server rejects the connection at the transport layer which surfaces as ECONNRESET rather than a clean "401 Unauthorized" error. This catches a lot of developers off guard because ECONNRESET looks like a network problem, not an auth problem.

The fix is a full re-authentication cycle:

# Step 1: Sign out completely
claude auth logout

# Step 2: Verify the old token is cleared
claude auth status
# Should show: "Not authenticated"

# Step 3: Sign back in — this fetches a fresh token
claude auth login

After re-authenticating, test with a simple prompt:

claude "say hello"

If this works, the auth token was the issue. If you still get ECONNRESET, the auth is fine and you're dealing with a network issue.

If You're Using an API Key Directly

Some developers configure Claude Code with a direct API key rather than OAuth login. Verify the key is set correctly and hasn't been revoked:

# Check current key configuration
echo $ANTHROPIC_API_KEY

# Should output your key starting with: sk-ant-api03-...
# If it's empty or shows a truncated/wrong value, reset it:
export ANTHROPIC_API_KEY="sk-ant-api03-YOUR-FULL-KEY-HERE"

You can verify and manage your API keys in the Anthropic Console. If a key has been revoked or regenerated, any Claude Code session using the old key will ECONNRESET. Our JWT decoder tool is useful for inspecting OAuth tokens paste the token to check its expiry timestamp and claims, all processed client-side in your browser.

Fix: Outdated Claude Code Version

Anthropic updates Claude Code's API compatibility alongside model releases. Older versions of Claude Code can have hardcoded API endpoint paths or authentication header formats that no longer work with the current API version the mismatch causes the server to reset the connection before completing the handshake.

# Check your current version
claude --version

# Update to latest
npm update -g @anthropic-ai/claude-code

# Or if you installed via a specific channel
npm install -g @anthropic-ai/claude-code@latest

After updating, run claude auth logout && claude auth login to ensure the new version initializes with a fresh token. Version mismatches and stale tokens together are especially likely to cause ECONNRESET fixing both at once is worth the 60 seconds.

Fix: Firewall and Network Path Issues

If the curl diagnostic returned a connection timeout not ECONNRESET, but a complete failure to connect your network is blocking Anthropic's API endpoints entirely rather than resetting active connections. This is a different problem with different fixes.

Check for Firewall Rules

# Test basic TCP connectivity to Anthropic's API
nc -zv api.anthropic.com 443

# Expected output if open:
# Connection to api.anthropic.com port 443 [tcp/https] succeeded!

# If you get "Connection refused" or timeout:
# Port 443 to api.anthropic.com is blocked on your network

If port 443 is blocked, you're on a restrictive network (some corporate networks, university networks, certain countries' ISPs) that requires either a VPN to bypass or explicit firewall rule changes.

DNS Resolution Check

Occasionally, DNS-level blocking prevents Claude Code from resolving Anthropic's domains at all:

# Check if DNS resolves correctly
nslookup api.anthropic.com

# Should return an IP address like:
# Address: 104.18.x.x or 162.159.x.x (Cloudflare-hosted)

# If it returns NXDOMAIN or a private IP:
# Your DNS is blocking or redirecting the domain

If DNS is the issue, switching to a public DNS resolver often fixes it:

# Temporarily test with Cloudflare DNS
curl --dns-servers 1.1.1.1 -v https://api.anthropic.com/v1/messages

Try a Different Network Path

ECONNRESET specifically (not timeout) can be path-specific your ISP's routing to Anthropic's servers has a bad hop that causes TCP resets on long-running connections. Mobile hotspot routes through a completely different network path and often resolves it instantly.

# On Mac/Linux — trace the network path to see where it breaks
traceroute api.anthropic.com

# On Windows
tracert api.anthropic.com

If the trace shows packet loss or timeouts at a specific hop, that's the problematic routing node. You can't fix it directly, but switching to a different network (hotspot, different WiFi) routes around it.

Fix: Claude Code Process Stuck in Bad State

Sometimes after a previous ECONNRESET, Claude Code's internal state gets stuck it keeps retrying with a bad connection object instead of creating a fresh one. A clean restart with process cleanup usually resolves this:

# Kill any lingering Claude Code processes
pkill -f "claude" 2>/dev/null

# On Windows PowerShell
Get-Process | Where-Object {$_.Name -like "*claude*"} | Stop-Process

# Clear Claude Code's temp connection cache
rm -rf ~/.claude/tmp/ 2>/dev/null

# Restart fresh
claude "test connection"

Also clear any environment variables that might be interfering with Claude Code's HTTP client configuration:

# Check for conflicting NODE_OPTIONS
echo $NODE_OPTIONS
echo $NODE_TLS_REJECT_UNAUTHORIZED

# If NODE_TLS_REJECT_UNAUTHORIZED is set to 0, reset it:
# (This setting disables TLS verification globally — dangerous and can cause ECONNRESET paradoxically)
unset NODE_TLS_REJECT_UNAUTHORIZED

When Nothing Works - Verify It's Actually Anthropic's Side

If you've worked through every fix above and still get ECONNRESET, the issue may genuinely be on Anthropic's infrastructure even if the official status page shows green. The official page has historically lagged real incidents by 30-120 minutes.

Check these three sources together:

• status.anthropic.com official, sometimes lags

• isdown.app/status/claude-ai user reports, often faster than official

• Twitter/X search "Claude Code ECONNRESET" real-time developer reports

If multiple developers are reporting the same issue simultaneously, it's infrastructure wait it out. The average Claude API incident resolution time based on May 2026 tracking data is approximately 30-60 minutes for API-specific issues.

For context on the broader May 2026 outage pattern and what Anthropic's infrastructure situation looks like right now, our Claude AI down and not working guide has the full incident timeline and status check workflow.

Fallback: Keep Working When Claude Code Is Down

The productive move when Claude Code is genuinely unavailable whether due to infrastructure or a networking issue you can't immediately resolve is having a local fallback ready. Running a local model with Ollama means you're never completely blocked:

# Install Ollama (one-time setup)
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a capable coding model
ollama pull llama3

# Run it locally — no internet required
ollama run llama3

It's not Claude-level quality for complex reasoning, but for code completion, debugging, and quick questions while waiting for Claude Code to recover, it's perfectly functional. Our complete Ollama + Next.js setup guide walks through integrating a local model into your development workflow so you always have an offline option ready the full setup takes about 10 minutes.

For developers building workflows that need to be resilient against any single AI provider going down, the Claude Code skills and complete guide covers building your Claude Code setup in a way that degrades gracefully to other providers when the primary connection fails.

Frequently Asked Questions

What does ECONNRESET mean in Claude Code?

ECONNRESET (Connection Reset) means the TCP connection between Claude Code and Anthropic's API server was established but then forcibly closed the server sent a reset packet before completing the response. Unlike a timeout (where no connection forms) or a 500 error (where the server responds with an error), ECONNRESET specifically means the connection was cut mid-handshake. In Claude Code's context, this is most commonly caused by VPN or proxy TLS inspection, an expired auth token, or a bad network routing path not usually an Anthropic infrastructure issue if the status page shows green.

Why does Claude Code show ECONNRESET when Anthropic's status page is green?

The status page showing green is actually a strong indicator that the problem is on your end, not Anthropic's. The three most likely causes in that scenario: your VPN or corporate proxy is intercepting the TLS connection and the server is resetting it, your Claude Code auth token has expired or corrupted and the server is rejecting the connection at the transport layer, or your ISP's routing path to Anthropic's servers has a bad hop causing TCP resets. The 30-second curl diagnostic in this guide tells you which of the three it is within seconds.

How do I fix Claude Code ECONNRESET on a corporate network?

Corporate networks run TLS inspection proxies that intercept HTTPS traffic this breaks Claude Code's TLS verification and causes ECONNRESET. The cleanest fix is asking your IT team to whitelist api.anthropic.com, claude.ai, and anthropic.com on the proxy inspection rule. While waiting for IT, test on mobile hotspot to confirm the corporate proxy is the cause and use hotspot as a working workaround. You can also try setting HTTPS_PROXY environment variable explicitly sometimes explicit proxy configuration resolves auto-detection issues that cause the reset.

Does re-authenticating Claude Code fix ECONNRESET?

Yes, in a significant number of cases. Run claude auth logout followed by claude auth login to regenerate your auth token completely. Expired or corrupted tokens cause the API server to reset the connection at the transport layer, which presents as ECONNRESET rather than a clean authentication error. This is counterintuitive most developers assume ECONNRESET is a network problem but auth token issues are one of the top three causes. Always test re-authentication before assuming it's a network issue.

Claude Code ECONNRESET only happens on WiFi but not mobile hotspot - why?

This is a network path issue. Your WiFi network (corporate, home, or public) takes a different routing path to Anthropic's API servers than your mobile carrier's network. If that WiFi path has a bad intermediate hop a router that sends TCP resets on long-lived HTTPS connections, or an ISP node with packet loss Claude Code will ECONNRESET on WiFi but work fine on mobile data. The fix is either using hotspot when on that network or working with your network admin to identify and resolve the bad routing hop.

How do I check if Claude Code ECONNRESET is Anthropic's fault or mine?

Run the curl diagnostic: curl -v 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-haiku-4-5-20251001","max_tokens":10,"messages":[{"role":"user","content":"hi"}]}'. If you get any HTTP response (even a 400), the API is reachable and the problem is in Claude Code's configuration. If you get a TLS error, it's a VPN or proxy issue. If you get a connection timeout, it's a firewall or routing issue. Only if you get ECONNRESET on the raw curl on multiple different networks is it likely an Anthropic-side infrastructure issue.

Can I use Claude Code without internet using a local model?

Claude Code itself requires Anthropic's API it cannot run completely offline. However, you can set up Ollama to run local models like Llama 3 or DeepSeek entirely on your machine with no internet required. It won't have Claude's reasoning quality for complex tasks, but it handles code completion, debugging questions, and routine development tasks well. Our Ollama + Next.js guide covers the full setup. Having this as a fallback means you're never completely blocked when Claude Code is unavailable whether due to ECONNRESET, outages, or network issues.