Databricks MCP Server

"""MCP Prompts loader - SIMPLE implementation per CLAUDE.md.""" import glob import os import re import yaml def load_prompts(mcp_server): """Load prompts from markdown files with MCP metadata. Parses YAML frontmatter for MCP configuration and registers prompts dynamically. """ prompt_files = glob.glob('prompts/*.md') for prompt_file in prompt_files: # Parse the markdown file for metadata and content metadata, content = parse_prompt_file(prompt_file) if not metadata: # Skip files without YAML frontmatter prompt_name = os.path.splitext(os.path.basename(prompt_file))[0] print(f'Warning: Skipping {prompt_name} - no YAML frontmatter found') continue # Register with full MCP metadata register_mcp_prompt(mcp_server, metadata, content) def parse_prompt_file(filepath): """Parse markdown file for YAML frontmatter and content.""" with open(filepath, 'r') as f: raw_content = f.read() # Check for YAML frontmatter (between --- markers) frontmatter_match = re.match(r'^---\s*\n(.*?)\n---\s*\n(.*)$', raw_content, re.DOTALL) if frontmatter_match: try: metadata = yaml.safe_load(frontmatter_match.group(1)) content = frontmatter_match.group(2) return metadata, content except yaml.YAMLError as e: print(f'Error parsing YAML in {filepath}: {e}') return None, raw_content return None, raw_content def register_mcp_prompt(mcp_server, metadata, content): """Register prompt with MCP metadata support. Note: FastMCP doesn't support argument validation in prompts yet, so we return the content with placeholder substitution but without runtime validation. The YAML metadata documents the expected arguments for future MCP compliance. """ name = metadata.get('name', 'unnamed_prompt') description = metadata.get('description', '') arguments = metadata.get('arguments', []) # Store metadata for documentation (even though FastMCP doesn't use it yet) # This prepares for future MCP compliance when FastMCP adds argument support # Create the prompt handler - FastMCP requires no parameters or specific named params @mcp_server.prompt(name=name, description=description) async def handle_prompt(): # Build comprehensive documentation including arguments text = content # Add argument documentation to the prompt if arguments: arg_docs = '\n\n## Expected Arguments:\n' for arg in arguments: required = ' (required)' if arg.get('required') else ' (optional)' arg_docs += f'- **{arg["name"]}**{required}: {arg.get("description", "No description")}\n' text = text + arg_docs return [{'role': 'user', 'content': {'type': 'text', 'text': text}}] # Note: No fallback support - all prompts must have YAML frontmatter # This ensures consistency and proper validation across all prompts