Model Context Protocol (MCP) and Autonomous Multi-Agent File Systems in Computer Science Education: Streamable HTTP Telemetry, Quarantine Security, and Compute Cashback Econometrics
Author: Julian Mercer (ORCID: 0009-0003-1162-170X) · Date: October 28, 2026 · DOI: 10.5281/zenodo.23756830
Abstract (ERIC 5-Element Standard)
The purpose of this study is to examine the architectural design, security telemetry, and pedagogical integration of remote Model Context Protocol (MCP) file servers for autonomous artificial intelligence agents in graduate and advanced undergraduate computer science curricula. Using a 180-run multi-agent benchmark dataset (DS-8) across six MCP client environments (Claude Desktop, Claude Code CLI, Cursor IDE, ChatGPT Apps, VS Code Copilot, and Gemini CLI), the methodology evaluates Streamable HTTP and local stdio transport latency, scoped auto-expiring link handoffs, The Armory pre-transit malware quarantine scanning, and Compute Cashback output reuse economics. Results show that direct agent-to-agent file exchange via LimeWire MCP eliminates the 'human proxy bottleneck,' completes cross-agent payload handoffs in 158 to 580 milliseconds, quarantines 100 percent of injected untrusted archive payloads prior to client disk write, and reduces redundant generative compute consumption by up to 75.0 percent through output reuse. Implications for AI engineering educators include a structured laboratory curriculum for multi-agent systems. Additional materials include 6 tables, 34 references, JSON-RPC schemas, and 195 embedded URI links.
Statutory ERIC Topic Areas & Descriptors
Topic Areas: Information and Technology; Science, Mathematics, and Environmental Education; Higher Education
ERIC Thesaurus Descriptors: Artificial Intelligence; Computer Science Education; Computer Networks; Information Security; Automation; Programming