MCP Meeting Agent

""" Tools package for LangChain tool definitions. """ import functools from typing import Callable, Any, Optional from ..core.logging_config import get_logger logger = get_logger(__name__) def tool_error_handler(fallback_prompt, hardcoded_fallback: str, model: str = "gpt-4o-mini"): """ Decorator to handle common error patterns in LangChain tools with LLM fallback. Args: fallback_prompt: The LangChain prompt template for LLM fallback hardcoded_fallback: Hardcoded fallback string if LLM fails model: The LLM model to use for fallback generation """ def decorator(func: Callable) -> Callable: @functools.wraps(func) async def wrapper(*args, **kwargs): try: return await func(*args, **kwargs) except Exception as e: logger.error(f"Error in {func.__name__}", error=str(e)) # Extract ctx from kwargs if present ctx = kwargs.get('ctx') # If we have MCP context, use LLM to generate contextual fallback if ctx: try: # Use MCP context to sample from LLM for dynamic fallback response = await ctx.sample( model=model, messages=fallback_prompt.format_messages(), max_tokens=200 ) return response.content except Exception as llm_error: logger.error(f"Error generating LLM fallback for {func.__name__}", error=str(llm_error)) # Fallback to hardcoded content if LLM generation fails return hardcoded_fallback return wrapper return decorator