Manage Multiple Claude Code Accounts.md

Manage Multiple Claude Code Accounts

Run two Claude Code accounts simultaneously on macOS without re-authenticating by using separate configuration directories.

Steps

1. Create Separate Config Directories

   mkdir ~/.claude-account1
   mkdir ~/.claude-account2

2. Add Aliases to Shell Config

   • Open ~/.zshrc (or ~/.bashrc for Bash) in an editor:

     nano ~/.zshrc

   • Add:

     alias claude-account1="CLAUDE_CONFIG_DIR=~/.claude-account1 claude"
     alias claude-account2="CLAUDE_CONFIG_DIR=~/.claude-account2 claude"

   • Save and reload:

     source ~/.zshrc

3. Authenticate Each Account

   • For account 1:

     claude-account1

     Log in; credentials save to ~/.claude-account1.
   • For account 2:

     claude-account2

     Log in; credentials save to ~/.claude-account2.

4. Switch or Run Simultaneously

   • Run claude-account1 or claude-account2 in separate terminal tabs (Cmd + T).
   • Each uses its own config and usage limits.

Notes

• Requires separate Anthropic accounts (different emails).
• Check Anthropic’s terms to ensure compliance.
• Update Claude Code: npm install -g @anthropic-ai/claude-code.
• If re-authentication occurs, verify CLAUDE_CONFIG_DIR with echo $CLAUDE_CONFIG_DIR.

While using multiple accounts I kept losing context when a session hit the 5-hour limit mid-flow. Built a hook that monitors
usage per account and silently warns Claude at 95% so it can save memory/notes before the session cuts off. Maybe it's useful to someone.

5. Monitor Usage Limits per Account

When running multiple accounts, you want to know when a specific account is approaching its 5-hour usage limit. This hook
automatically warns Claude at 95% usage — per account — using CLAUDE_CONFIG_DIR to target the right credentials.

Create the hook script (repeat for ~/.claude-account2/hooks/):

mkdir -p ~/.claude-account1/hooks
nano ~/.claude-account1/hooks/session-age-warn.sh

Paste:

#!/bin/bash
# Warn when Claude Pro 5-hour usage hits 95%.
# Runs as a UserPromptSubmit hook; receives JSON payload on stdin.

cat > /dev/null  # drain stdin

# Derive keychain service name from config dir (sha256 prefix, same as Claude Code)
config_dir="${CLAUDE_CONFIG_DIR:-$HOME/.claude}"
hash=$(python3 -c "import hashlib,sys; print(hashlib.sha256(sys.argv[1].encode()).hexdigest()[:8])" "$config_dir" 2>/dev/null)
svc="Claude Code-credentials-${hash}"

creds=$(security find-generic-password -s "$svc" -w 2>/dev/null)
if [ -z "$creds" ]; then exit 0; fi

TOKEN=$(echo "$creds" | python3 -c "
import sys, json, time
d = json.load(sys.stdin)
oauth = d.get('claudeAiOauth', {})
expires = oauth.get('expiresAt', 0)
token = oauth.get('accessToken', '')
if token and int(time.time()*1000) < expires:
    print(token)
" 2>/dev/null)

if [ -z "$TOKEN" ]; then exit 0; fi

response=$(curl -s "https://api.anthropic.com/api/oauth/usage" \
  -H "Authorization: Bearer $TOKEN" \
  -H "anthropic-beta: oauth-2025-04-20" \
  -H "User-Agent: claude-code/2.0.32" 2>/dev/null)

utilization=$(echo "$response" | python3 -c "import sys,json; d=json.load(sys.stdin);
print(int(d['five_hour']['utilization']))" 2>/dev/null)

if [ -z "$utilization" ]; then exit 0; fi

if [ "$utilization" -ge 95 ]; then
  echo "{\"hookSpecificOutput\": {\"hookEventName\": \"UserPromptSubmit\", \"additionalContext\": \"[5-hour usage at
${utilization}% — update memory files before this session ends.]\"}}"
fi

chmod +x ~/.claude-account1/hooks/session-age-warn.sh

Register the hook in ~/.claude-account1/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "/Users/YOUR_USERNAME/.claude-account1/hooks/session-age-warn.sh"
          }
        ]
      }
    ]
  }
}

Repeat for ~/.claude-account2/, pointing the command path to that directory.

How it works:

• Claude Code stores credentials in macOS Keychain under a key derived from CLAUDE_CONFIG_DIR (sha256 prefix). The hook uses
  the same derivation to look up the right account's token.
• It calls https://api.anthropic.com/api/oauth/usage and checks five_hour.utilization.
• At 95%+, it injects a silent context warning that Claude sees on your next message, prompting it to save memory/notes before
  the session ends.
• Silently skips if the token is expired or the API is unreachable.