Session Token Usage

Extract and report input/output token counts from Claude Code session logs.

Steps

1. Identify session

project_dir="$HOME/.claude/projects/$(pwd | sed 's|/|-|g')"
current_session=$(ls -t "$project_dir"/*.jsonl 2>/dev/null | head -1)
session_id=$(basename "$current_session" .jsonl)
echo "Session: $session_id"
echo "Source: $current_session"

If the user specifies a session ID, use that instead of the most recent one.

2. Extract token usage

python3 -c "
import json, sys

log_file = sys.argv[1]

# Deduplicate by message ID — session logs may contain multiple
# JSONL lines per API call (streaming). Take the last usage block
# per unique message ID for accurate counts.
usage_by_msg = {}

with open(log_file) as f:
    for line in f:
        line = line.strip()
        if not line:
            continue
        try:
            obj = json.loads(line)
        except json.JSONDecodeError:
            continue
        msg = obj.get('message', {})
        msg_id = msg.get('id')
        usage = msg.get('usage')
        if msg_id and usage:
            usage_by_msg[msg_id] = usage

totals = {
    'input_tokens': 0,
    'output_tokens': 0,
    'cache_creation_input_tokens': 0,
    'cache_read_input_tokens': 0,
}

for usage in usage_by_msg.values():
    for key in totals:
        totals[key] += usage.get(key, 0)

msg_count = len(usage_by_msg)
total_input = totals['input_tokens'] + totals['cache_creation_input_tokens'] + totals['cache_read_input_tokens']
print(f'API calls:                   {msg_count:>12,}')
print(f'Input tokens:                {totals[\"input_tokens\"]:>12,}')
print(f'Output tokens:               {totals[\"output_tokens\"]:>12,}')
print(f'Cache creation input tokens: {totals[\"cache_creation_input_tokens\"]:>12,}')
print(f'Cache read input tokens:     {totals[\"cache_read_input_tokens\"]:>12,}')
print(f'Total input (all types):     {total_input:>12,}')
" "$current_session"

3. Report results

Present results as a clean table:

• Session ID — the UUID
• API calls — number of unique API calls with usage data
• Input tokens — non-cached input tokens
• Output tokens — tokens generated by the model
• Cache creation tokens — input tokens written to prompt cache
• Cache read tokens — input tokens served from prompt cache
• Total input — sum of all input token types