Keep-Alive Timeout Mismatch Error — Fix (2026)

The Error

Error: socket hang up
  code: 'ECONNRESET'
  at connResetException (node:internal/errors:720:14)
  Reusing keep-alive connection that was closed by server

This error occurs when the client reuses a TCP connection that the server has already closed due to keep-alive timeout expiration. The client thinks the connection is live but sends data to a closed socket.

The Fix

  1. Disable keep-alive to force fresh connections:
export ANTHROPIC_DISABLE_KEEPALIVE=1
claude "test"
  1. In the Node.js SDK, configure the HTTP agent:
import { Agent } from 'node:https';
const agent = new Agent({
  keepAlive: true,
  keepAliveMsecs: 15000, // Less than server timeout
  maxSockets: 10
});
  1. Or disable keep-alive entirely in the SDK:
import { Agent } from 'node:https';
const agent = new Agent({ keepAlive: false });

Why This Happens

HTTP keep-alive allows multiple requests over a single TCP connection. The server closes idle connections after a timeout (typically 30-60 seconds). If the client’s keep-alive timeout is longer than the server’s, the client may try to send a request on a connection that the server has already closed. The server sends a RST packet, causing “socket hang up.”

If That Doesn’t Work

  • Add retry logic specifically for ECONNRESET on reused connections:
import anthropic
client = anthropic.Anthropic(max_retries=3)
# SDK automatically retries on ECONNRESET
  • Check if a proxy modifies keep-alive headers:
curl -v -H "Connection: keep-alive" https://api.anthropic.com 2>&1 | grep -i "keep-alive\|connection"
  • Reduce the max idle time on your HTTP client to below 30 seconds.

Prevention

Add this to your CLAUDE.md:

# Connection Management
- Set keep-alive timeout to 15s (less than server's 30s default).
- Always configure max_retries for ECONNRESET recovery.
- In batch processing, disable keep-alive to avoid stale connections.

This fix also applies if you see these related error messages:

  • ETIMEDOUT: connection timed out
  • RequestTimeout: request took longer than 120000ms
  • ESOCKETTIMEDOUT
  • ECONNREFUSED: connection refused through proxy
  • Error: unable to verify the first certificate

Frequently Asked Questions

What is the default timeout for Claude Code API requests?

The default timeout is 120 seconds (120000ms). For complex operations involving large codebases or multi-file edits, this may be insufficient. Increase it with claude config set api_timeout 300000 for a 5-minute timeout.

Can network latency cause timeouts?

Yes. Corporate proxies, VPNs, and DNS filtering services add round-trip latency. Measure your baseline latency with curl -o /dev/null -s -w '%{time_total}' https://api.anthropic.com/v1/messages. If it exceeds 5 seconds, route API traffic outside the proxy.

Do timeouts consume API credits?

Partially. If the server began processing your request before the client timed out, the input tokens are consumed even though you never received a response. Long timeouts reduce wasted credits by allowing the response to complete.

How do I configure Claude Code to use a corporate proxy?

Set the HTTPS_PROXY environment variable: export HTTPS_PROXY=http://proxy.corp.com:8080. Claude Code respects standard proxy environment variables. Add this to your shell profile for persistence.

Configure MCP → Build your server config with our MCP Config Generator.

Try it: Paste your error into our Error Diagnostic for an instant fix.

Timeout Configuration Reference

Claude Code has several timeout settings that interact with each other:

Setting Default Controls How to Change
Bash timeout 120s Maximum time for a single bash command CLAUDE_CODE_BASH_TIMEOUT=600
API timeout 300s Maximum time waiting for API response Network-level setting
Session timeout None Auto-close after inactivity Not currently configurable
MCP server init 30s Time allowed for MCP server startup Set in settings.json

Diagnosing Slow Operations

When Claude Code appears to hang, determine which component is slow:

Step 1: Check CPU and memory. Run top -l 1 | grep -E "node|claude" (macOS) or top -bn1 | grep -E "node|claude" (Linux). High CPU suggests active computation. High memory (over 1GB) suggests the conversation context is too large.

Step 2: Check network connectivity. Run curl -s -o /dev/null -w "%{time_total}" https://api.anthropic.com/v1/messages. Response times over 2 seconds indicate network issues between you and the API.

Step 3: Check disk I/O. Run iostat 1 3 to see if disk is a bottleneck. Claude Code performs significant file reads when scanning large projects. An SSD reduces file scanning from minutes to seconds.

Step 4: Reduce context size. If the session has been running for many turns, accumulated context can slow API responses. Use /clear to reset the conversation and start fresh with a smaller context window.