PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
Synapse
⌘+K
Synapse
March 1, 2026Open Access

Pylon-7: A 7-Layer Reference Model for AI Agent Workflows -- Exploratory Study on MCP Layering for Efficiency, Accuracy, and Safety in Commodity Hardware Environments

View Full Paper
Ask AI
Bookmark
Share

Authors

HJHyunwoo Jeon

Discussion

Loading...

Member takes

Overview

Exploratory work explores how MCP layering improves efficiency, accuracy, and safety in AI agents using commodity hardware.

Key Points

  • This research aims to establish a systematic reference model for AI agent workflows using Pylon-7.
  • Proposed a 7-layer reference model for AI agents inspired by the OSI model.
  • Conducted 615 experimental runs across 10 scenarios using Qwen 2.5 and GPT-OSS 20B models.
  • Explored efficiency, accuracy, and safety at 5 MCP depth levels (L0-L4) on CPU-only systems.
  • MCP layering reduced token usage by 47% while increasing accuracy by 37%.
  • Layer 3 (L3) provided optimal structured output and candidate actions.
  • A 7B model combined with MCP was 3.5x cheaper and 14.4% more accurate than using a 20B model alone.
  • Tasks previously deemed impossible achieved perfect performance with the MCP framework.
  • Above L3, both models maintained an accuracy of 0.93+, indicating MCP's structural advantages.

Cite This Study

Hyunwoo Jeon (2026) studied this question.

synapsesocial.com/papers/69a3d856ec16d51705d2f135https://doi.org/10.5281/zenodo.18808597
View Full Paper
Ask AI
Bookmark
Share