#!/usr/bin/env python3
"""
Prepare data files from the reference code for the MCP server.
This script extracts the few-shot examples and schema from the reference directory.
"""
import json
import shutil
from pathlib import Path
# Paths
REFERENCE_DIR = Path(__file__).parent.parent / "reference"
DATA_DIR = Path(__file__).parent
def prepare_few_shot_examples():
"""Extract and convert few-shot examples from reference format to MCP format.
Creates 132 individual examples (66 configs × 2 descriptions each):
- 66 with human-written descriptions
- 66 with GPT-4 generated descriptions
"""
print("📦 Preparing few-shot examples...")
# Load the reference data
source_file = REFERENCE_DIR / "all_few_shots.json"
with open(source_file) as f:
data = json.load(f)
# Extract the Questions (which are the few-shot examples)
# Create TWO separate examples for each: one with human description, one with GPT-4 description
examples = []
example_id = 0
for i, item in enumerate(data["Questions"]):
# Each item has: Type, Answer (config), Question, QuestionGPT4
# Example 1: Human-written description
example_human = {
"id": example_id,
"type": item["Type"],
"description": item["Question"], # Original human description
"config": item["Answer"],
"headers": ",".join(data["data"][0]), # Column headers from data
"source": "human",
"original_id": i
}
examples.append(example_human)
example_id += 1
# Example 2: GPT-4 generated description (for same config)
example_gpt4 = {
"id": example_id,
"type": item["Type"],
"description": item["QuestionGPT4"], # GPT-4 generated alternative
"config": item["Answer"],
"headers": ",".join(data["data"][0]),
"source": "gpt4",
"original_id": i
}
examples.append(example_gpt4)
example_id += 1
# Save in our format
output_file = DATA_DIR / "few_shot_examples.json"
with open(output_file, "w") as f:
json.dump(examples, f, indent=2)
print(f"✓ Created {len(examples)} few-shot examples in {output_file}")
print(f" - {len(examples)//2} human-written descriptions")
print(f" - {len(examples)//2} GPT-4 generated descriptions")
return len(examples)
def copy_schema():
"""Copy schema file from reference."""
print("📦 Copying schema...")
source_file = REFERENCE_DIR / "schema.txt"
dest_file = DATA_DIR / "schema.md"
shutil.copy(source_file, dest_file)
print(f"✓ Copied schema to {dest_file}")
def copy_prompt_template():
"""Copy prompt template from reference."""
print("📦 Copying prompt template...")
source_file = REFERENCE_DIR / "prompt.md"
dest_file = DATA_DIR / "prompt_template.md"
shutil.copy(source_file, dest_file)
print(f"✓ Copied prompt template to {dest_file}")
def main():
"""Prepare all data files."""
print("=" * 60)
print("Preparing CanvasXpress MCP Server Data")
print("=" * 60)
# Create data directory if it doesn't exist
DATA_DIR.mkdir(exist_ok=True)
# Prepare each component
num_examples = prepare_few_shot_examples()
copy_schema()
copy_prompt_template()
print("=" * 60)
print(f"✅ Data preparation complete!")
print(f" - {num_examples} few-shot examples")
print(f" - Schema documentation")
print(f" - Prompt template")
print("=" * 60)
if __name__ == "__main__":
main()
