ARTFEED — Contemporary Art Intelligence

ColPackAgent: AI Agent for Colloidal Packing Simulations

ai-technology · 2026-05-18

ColPackAgent serves as an agent framework designed to independently conduct Monte Carlo simulations focused on colloidal packing. It utilizes a Model Context Protocol (MCP) tool server along with an agent skill to carry out organized workflows. By integrating HOOMD-blue hard-particle Monte Carlo through a specialized colpack Python package, the system can function interactively with human input, autonomously based on prompts, or in an autoresearch mode. This development tackles the shortcomings of general-purpose LLM agents, which tend to describe workflows instead of executing them with consistency.

Key facts

  • ColPackAgent is an agent framework for Monte Carlo simulations of colloidal packing.
  • It uses a Model Context Protocol (MCP) tool server and an agent skill.
  • The MCP server exposes a custom colpack Python package wrapping HOOMD-blue hard-particle Monte Carlo.
  • The skill encodes a four-stage workflow contract.
  • ColPackAgent can operate interactively, autonomously, or as autoresearch.
  • General-purpose LLM agents tend to describe workflows rather than execute them reliably.
  • The framework is central to studies of phase behavior, self-assembly, and materials design.
  • The paper is available on arXiv with ID 2605.15625.

Entities

Institutions

  • arXiv

Sources