RAG 与 MCP：为企业做出正确的选择.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.DEV342RAG vs MCP:Making the right choice for enterprisesPooja RanjanLead Data ScientistIQMFernando PenaHead of Cloud ServicesNumen 2025,Amazon Web Services,Inc.or it

2、s affiliates.All rights reserved.75%Filtering of hallucinated responses from models for RAG and summarization use casesRAG vs MCPWhy does it matter?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaWhy does it

3、matter?Definition:RAG and MCPThe Choices:RAG and MCPRAG and MCP on AWSRAG vs MCP:Side-by-SideDecision FrameworkTakeaways 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Why does it matter?Generative AI adoption is acceleratingKey architectural choice:RAG vs MCPRisks:complexity,cos

4、t,scalability,securityGoal provide a decision framework 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.RAG and MCP:Choices?2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Definition:RAG and MCPo

5、Enhances LLMs with enterprise dataoUses documents,databases,and knowledge bases to ground responsesoReduces hallucinations and improves accuracyoKeeps answers aligned with enterprise-specific informationoBest for search,Q&A,and knowledge-heavy tasksRAG(Retrieval-Augmented Generation)MCP(Model Contex

6、t Protocol)oOpen protocol for connecting LLMs to tools,APIs,and systemsoAllows models to take actions,not just read informationoEnables querying structured data,executing workflows,or running codeoFoundation for agentic and spec-driven AI applicationsoBest for automation,system integration,and tool-