MCP Servers in Agentic AI - Architecture, Adoption, and Security of the Model Context Protocol
Year: 2026
Type: preprint
Abstract: The Model Context Protocol (MCP), introduced by Anthropic in late 2024, has emerged as a widely adopted open standard forconnecting large language model (LLM) applications to external tools, data sources, and services. By standardizing the interfacebetween an AI host application and the servers that expose capabilities to it, MCP has been described as a USB-C-like connectorfor agentic AI, replacing bespoke, one-off integrations with a single, reusable protocol. This paper explains how MCP servers work within agentic AI systems: the host-client-server architecture, the JSON-RPC 2.0 messaging layer, the core primitives of tools, resources, and prompts, and the rapid evolution of the specification from an initial local-only design to a stateless, horizontally scalable protocol core in its 2026-07-28 revision. We review the growing ecosystem of MCP servers and the parallel emergence of complementary agent-to-agent protocols, synthesize empirical security research documenting tool poisoning, indirect prompt injection, and auto-execution vulnerabilities across production AI coding tools, and summarize benchmarking efforts that evaluate LLM agents against real-world MCP server environments. We conclude with governance guidance for deploying MCP servers safely in production agentic systems and an agenda for future research.
Author: Rajagopal, Deepthi
Institution: Binghamton University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7208735451