#!/usr/bin/env python3
"""
Demonstrate how MCP enables precise targeted edits vs full file rewrites
"""
import os
import sys
import sqlite3
import json
from pathlib import Path
# Add parent directory for imports
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from mcp_server.utils.token_counter import TokenCounter
def search_for_edit_target(db_path: str, query: str):
"""Search for code to edit using MCP."""
conn = sqlite3.connect(db_path)
cursor = conn.cursor()
# Search for specific function/symbol
cursor.execute(
"""SELECT
filepath,
snippet(bm25_content, 3, '<<<', '>>>', '...', 30) as snippet,
bm25(bm25_content) as score
FROM bm25_content
WHERE bm25_content MATCH ?
ORDER BY score
LIMIT 5""",
(query,)
)
results = cursor.fetchall()
conn.close()
return [(r[0], r[1]) for r in results]
def demonstrate_edit_scenarios():
"""Show different edit scenarios with MCP vs traditional search."""
token_counter = TokenCounter()
print("šÆ MCP Edit Precision Demonstration")
print("="*60)
# Scenario 1: Fix a specific function
print("\nš Scenario 1: Fix error handling in Gin middleware")
print("-"*60)
# With MCP
print("\nš With MCP:")
mcp_results = search_for_edit_target('.mcp-index/gin/code_index.db', 'func Recovery')
if mcp_results:
file_path, snippet = mcp_results[0]
print(f"1. MCP search for 'func Recovery': Found in {file_path}")
print(f" Snippet: {snippet[:150]}...")
print(f"2. MCP provides exact location ā Use Edit tool for targeted change")
print(f"3. Only modified lines are changed (diff-based edit)")
# Simulate edit
edit_example = {
'tool': 'Edit',
'file_path': file_path,
'old_string': 'panic(err)',
'new_string': 'c.AbortWithError(500, err)'
}
edit_tokens = token_counter.count_tokens(json.dumps(edit_example))
print(f" Edit tokens: ~{edit_tokens}")
print("\nā Without MCP:")
print("1. Grep search finds multiple files")
print("2. Must read entire files to understand context")
print("3. Often results in Write tool usage (full file rewrite)")
print(" Full file tokens: ~5000-10000")
# Scenario 2: Add logging to specific handler
print("\n\nš Scenario 2: Add logging to specific API handler")
print("-"*60)
print("\nš With MCP:")
mcp_results = search_for_edit_target('.mcp-index/gin/code_index.db', 'Handler context')
if mcp_results:
file_path, snippet = mcp_results[0]
print(f"1. MCP search for handler functions: Found in {file_path}")
print(f" Snippet: {snippet[:150]}...")
print(f"2. Precise location enables MultiEdit for multiple small changes")
print(f"3. Each edit is a small targeted diff")
print(f" MultiEdit tokens: ~200-300 per edit")
print("\nā Without MCP:")
print("1. Broad search returns many handlers")
print("2. Manual inspection needed to find right one")
print("3. Risk of modifying wrong handler")
print("4. Full file rewrite more likely")
# Scenario 3: Update import statements
print("\n\nš Scenario 3: Update import path across codebase")
print("-"*60)
print("\nš With MCP:")
mcp_results = search_for_edit_target('.mcp-index/react/code_index.db', 'import from react')
if mcp_results[:3]:
print(f"1. MCP finds exact import lines in {len(mcp_results)} files")
for i, (file_path, snippet) in enumerate(mcp_results[:3]):
print(f" File {i+1}: {file_path}")
print(f" Import: {snippet[:80]}...")
print(f"2. Can use Edit tool with replace_all=True for each file")
print(f"3. Only import lines are modified")
print(f" Edit tokens per file: ~50-100")
print("\nā Without MCP:")
print("1. Must read entire files to find imports")
print("2. Risk of incorrect regex replacements")
print("3. Often leads to full file rewrites")
print(" Full file tokens per file: ~5000+")
# Summary
print("\n\nš Summary: MCP vs Traditional Search for Code Edits")
print("="*60)
print("\nā
MCP Advantages:")
print("- Precise location finding (file + line + context)")
print("- Enables Edit/MultiEdit tools (diff-based changes)")
print("- 90-99% fewer tokens per edit operation")
print("- Lower risk of unintended changes")
print("- Faster iteration cycles")
print("\nā Traditional Search Limitations:")
print("- Must read full files for context")
print("- Often results in Write tool (full file replacement)")
print("- 10-100x more tokens per edit")
print("- Higher risk of breaking unrelated code")
print("- Slower, more error-prone process")
print("\nš” Key Insight:")
print("MCP transforms code editing from 'find and rewrite' to 'locate and patch'")
print("This enables Claude Code to make surgical edits instead of wholesale replacements")
if __name__ == "__main__":
demonstrate_edit_scenarios()
