src/mcp_core/processor.py

"""
MCP Core - Interaction Processor

This module provides the core functionality for processing interactions
between users and models through the Model Context Protocol.
"""

from typing import Dict, Any, Optional, List, Union
import asyncio

from .context import Context, ContextManager
from .model_adapters import ModelAdapter, ModelRegistry
from .protocol import MCPRequest, MCPResponse, InteractionType


class InteractionProcessor:
    """
    Processes interactions between users and models.
    
    This class handles the flow of interactions through the MCP system,
    managing context, routing to appropriate models, and formatting responses.
    """
    
    def __init__(
        self, 
        context_manager: ContextManager,
        model_registry: ModelRegistry
    ):
        """
        Initialize a new InteractionProcessor.
        
        Args:
            context_manager: The ContextManager to use for context operations.
            model_registry: The ModelRegistry to use for model operations.
        """
        self.context_manager = context_manager
        self.model_registry = model_registry
        
    async def process_interaction(
        self, 
        request: MCPRequest,
        model_name: str
    ) -> MCPResponse:
        """
        Process an interaction request.
        
        Args:
            request: The MCPRequest to process.
            model_name: Name of the model adapter to use.
            
        Returns:
            An MCPResponse containing the model's response.
            
        Raises:
            ValueError: If the context ID is invalid or the model is not found.
        """
        # Get or create context
        context_id = request.context_id
        context = None
        
        if context_id:
            # Use existing context
            context = self.context_manager.get_context(context_id)
            if not context:
                raise ValueError(f"Context with ID {context_id} not found")
        else:
            # Create new context
            context = self.context_manager.create_context(
                metadata={"interaction_type": request.interaction_type}
            )
            context_id = context.context_id
            
        # Get model adapter
        model_adapter = self.model_registry.get_adapter(model_name)
        if not model_adapter:
            raise ValueError(f"Model adapter '{model_name}' not found")
            
        # Process the request with the model
        model_input = {
            "text": request.content.get("text", ""),
            "interaction_type": request.interaction_type,
            "format": request.format,
            **request.metadata
        }
        
        # Include context state in model input
        model_context = {
            "state": context.state,
            "history": context.get_history(5)  # Last 5 interactions
        }
        
        # Process with model
        model_output = await model_adapter.process(model_input, model_context)
        
        # Update context with this interaction
        context.add_interaction(request.dict(), model_output)
        
        # Extract educational metadata if available
        educational_metadata = model_output.get("educational_metadata", {})
        if educational_metadata:
            # Update context state with educational information
            context.update_state({
                "educational": {
                    "last_topics": educational_metadata.get("topics", []),
                    "complexity_level": educational_metadata.get("complexity_level", ""),
                    "suggested_follow_ups": educational_metadata.get("suggested_follow_ups", [])
                }
            })
        
        # Create response
        response = MCPResponse(
            context_id=context_id,
            interaction_type=request.interaction_type,
            content={
                "text": model_output.get("text", ""),
                "educational_metadata": educational_metadata
            },
            format=request.format,
            metadata=request.metadata,
            model_info={
                "model_id": model_output.get("model_id", model_name),
                "confidence": model_output.get("confidence", 1.0)
            }
        )
        
        return response


class BatchProcessor:
    """
    Processes batches of interactions.
    
    This class provides functionality for processing multiple interactions
    in parallel or sequence.
    """
    
    def __init__(self, interaction_processor: InteractionProcessor):
        """
        Initialize a new BatchProcessor.
        
        Args:
            interaction_processor: The InteractionProcessor to use.
        """
        self.interaction_processor = interaction_processor
        
    async def process_batch(
        self, 
        requests: List[Dict[str, Any]],
        model_name: str,
        parallel: bool = True
    ) -> List[Dict[str, Any]]:
        """
        Process a batch of interaction requests.
        
        Args:
            requests: List of request dictionaries.
            model_name: Name of the model adapter to use.
            parallel: Whether to process requests in parallel.
            
        Returns:
            List of response dictionaries.
        """
        if parallel:
            # Process requests in parallel
            tasks = [
                self.interaction_processor.process_interaction(
                    MCPRequest(**request),
                    model_name
                )
                for request in requests
            ]
            responses = await asyncio.gather(*tasks)
        else:
            # Process requests sequentially
            responses = []
            for request in requests:
                response = await self.interaction_processor.process_interaction(
                    MCPRequest(**request),
                    model_name
                )
                responses.append(response)
                
        # Convert responses to dictionaries
        return [response.dict() for response in responses]