4.1.3. Model Context Protocol (MCP)

💡 First Principle: MCP (Model Context Protocol) standardizes the interface between AI models and external tools/data sources using a client-server architecture — instead of every agent framework implementing custom tool integration code, any MCP-compatible tool can be used with any MCP-compatible agent without modification. It is the USB standard for AI tool connectivity.

MCP is an open protocol developed by Anthropic and adopted by AWS in 2025. Understanding it is differentiating on the AIP-C01 exam because most preparation materials don't cover it. Exam scenarios involving tool integration across multiple agent frameworks or standardized API patterns for retrieval augmentation typically point to MCP.

MCP architecture:

MCP server implementation patterns:

Lambda-based MCP server — the exam-relevant pattern:

def lambda_handler(event, context):
    method = event.get('method')
    
    if method == 'initialize':
        return {
            'protocolVersion': '2024-11-05',
            'serverInfo': {'name': 'order-management-mcp', 'version': '1.0.0'},
            'capabilities': {'tools': {}}
        }
    
    elif method == 'tools/list':
        return {
            'tools': [
                {
                    'name': 'get_order_status',
                    'description': 'Retrieve current status and details for a customer order',
                    'inputSchema': {
                        'type': 'object',
                        'properties': {'order_id': {'type': 'string', 'description': 'The order ID'}},
                        'required': ['order_id']
                    }
                }
            ]
        }
    
    elif method == 'tools/call':
        tool_name = event['params']['name']
        args = event['params']['arguments']
        if tool_name == 'get_order_status':
            result = fetch_order_from_dynamodb(args['order_id'])
            return {'content': [{'type': 'text', 'text': json.dumps(result)}]}

MCP for vector store access (Skill 1.5.6): MCP provides a standardized function-calling interface for vector search queries, making retrieval components interchangeable across different agent implementations. An MCP-wrapped vector store can be queried identically by Bedrock Agents, Strands Agents, or any other MCP-compatible client.

⚠️ Exam Trap: MCP is not an AWS-proprietary standard — it's an open protocol. Exam scenarios that ask about "standardized tool integration across multiple agent frameworks" or "portable AI tool interfaces" point to MCP, not to a custom API Gateway + Lambda pattern. Recognizing MCP as the answer to interoperability problems is a key differentiator for passing this exam.

Reflection Question: Your organization runs three different AI agent frameworks: Bedrock Agents for customer service, Strands Agents for internal tools, and a third-party framework for data analysis. You have 15 internal data tools (inventory lookup, CRM access, calendar integration). Without MCP, how many custom integrations do you need? With MCP, how many? What does this illustrate about MCP's architectural value?