#!/usr/bin/env python3
"""
Simple analysis of Claude Code transcripts to extract real tool usage data.
"""
import json
import re
from pathlib import Path
from collections import defaultdict
def analyze_transcript(file_path):
"""Analyze a single transcript for tool usage."""
tool_sequences = []
current_sequence = []
read_details = []
grep_details = []
mcp_details = []
with open(file_path, 'r') as f:
for line in f:
try:
entry = json.loads(line.strip())
# Look for tool uses in assistant messages
if entry.get('type') == 'assistant' and entry.get('message'):
content_items = entry['message'].get('content', [])
for item in content_items:
if item.get('type') == 'tool_use':
tool_name = item.get('name', '')
tool_input = item.get('input', {})
tool_info = {
'name': tool_name,
'params': tool_input
}
current_sequence.append(tool_info)
# Collect specific details
if tool_name == 'Read':
read_details.append({
'file': tool_input.get('file_path', ''),
'offset': tool_input.get('offset', 0),
'limit': tool_input.get('limit', 'full')
})
elif tool_name == 'Grep':
grep_details.append({
'pattern': tool_input.get('pattern', ''),
'path': tool_input.get('path', '.')
})
elif tool_name.startswith('mcp__'):
mcp_details.append(tool_info)
# Look for tool results to estimate tokens
if entry.get('type') == 'user' and entry.get('toolUseResult'):
result = entry.get('toolUseResult', {})
# Track result sizes
if isinstance(result, dict):
if result.get('type') == 'text':
if result.get('file'):
# Read result
lines = result['file'].get('numLines', 0)
if current_sequence and current_sequence[-1]['name'] == 'Read':
current_sequence[-1]['result_lines'] = lines
elif result.get('filenames'):
# Grep result
num_files = result.get('numFiles', 0)
if current_sequence and current_sequence[-1]['name'] == 'Grep':
current_sequence[-1]['result_files'] = num_files
# New user message starts new sequence
if entry.get('type') == 'user' and entry.get('userType') == 'external':
if current_sequence:
tool_sequences.append(current_sequence)
current_sequence = []
except json.JSONDecodeError:
continue
# Add final sequence
if current_sequence:
tool_sequences.append(current_sequence)
return {
'sequences': tool_sequences,
'read_details': read_details,
'grep_details': grep_details,
'mcp_details': mcp_details
}
def main():
"""Analyze all transcripts in the project."""
project_dir = Path.home() / ".claude/projects/-workspaces-Code-Index-MCP"
all_reads = []
all_greps = []
all_mcps = []
all_patterns = []
# Analyze each transcript
for transcript_file in sorted(project_dir.glob("*.jsonl")):
print(f"\nAnalyzing {transcript_file.name}...")
result = analyze_transcript(transcript_file)
all_reads.extend(result['read_details'])
all_greps.extend(result['grep_details'])
all_mcps.extend(result['mcp_details'])
# Extract patterns
for sequence in result['sequences']:
if len(sequence) > 0:
pattern = ' -> '.join([t['name'] for t in sequence])
all_patterns.append(pattern)
# Print summary
print("\n" + "=" * 80)
print("REAL CLAUDE CODE TOOL USAGE ANALYSIS")
print("=" * 80)
print(f"\nTotal tool calls analyzed:")
print(f" Read: {len(all_reads)}")
print(f" Grep: {len(all_greps)}")
print(f" MCP: {len(all_mcps)}")
# Analyze Read patterns
if all_reads:
print(f"\nRead tool usage:")
limits = [r['limit'] for r in all_reads if r['limit'] != 'full']
if limits:
avg_limit = sum(limits) / len(limits)
print(f" Reads with limit: {len(limits)} ({len(limits)/len(all_reads)*100:.1f}%)")
print(f" Average limit: {avg_limit:.0f} lines")
print(f" Common limits: {sorted(set(limits))[:10]}")
full_reads = len([r for r in all_reads if r['limit'] == 'full'])
print(f" Full file reads: {full_reads} ({full_reads/len(all_reads)*100:.1f}%)")
# Analyze Grep patterns
if all_greps:
print(f"\nGrep tool usage:")
print(f" Total grep searches: {len(all_greps)}")
print(f" Example patterns: {[g['pattern'] for g in all_greps[:5]]}")
# Analyze tool sequences
if all_patterns:
print(f"\nCommon tool sequences:")
pattern_counts = defaultdict(int)
for p in all_patterns:
pattern_counts[p] += 1
for pattern, count in sorted(pattern_counts.items(), key=lambda x: x[1], reverse=True)[:10]:
print(f" {pattern}: {count}")
# MCP adoption
total_tools = len(all_reads) + len(all_greps) + len(all_mcps)
if total_tools > 0:
mcp_rate = len(all_mcps) / total_tools * 100
print(f"\nMCP Adoption Rate: {mcp_rate:.2f}%")
# Look for grep->read patterns
grep_to_read = 0
for pattern in all_patterns:
if 'Grep -> Read' in pattern:
grep_to_read += 1
print(f"\nGrep followed by Read: {grep_to_read} occurrences")
if __name__ == "__main__":
main()
