Aspen Plus模拟控制工具

Aspen Co-pilot

English · 繁體中文 · 简体中文 · 日本語 · 한국어 · Español · Português · Deutsch · Français · Русский

Let an AI agent design, simulate and optimize chemical processes in your own Aspen Plus — while you see every step and stay in charge.

Aspen Co-pilot connects Aspen Plus to AI agents such as Claude Code and Codex through the Model Context Protocol (MCP). You describe the process in plain language; the agent does the work in Aspen, and you review it.

Windows only. You need your own licensed Aspen Plus; this project does not include it. Developed by NCKU GESE Lab.

What it can do

Build and change processes. Create a flowsheet from a description — components, property methods, reactions, unit operations and streams — set and adjust any parameter, run the simulation and read the results. Most Aspen Plus functions are available.

Go from literature to a working model. With your AI client's research tools, the agent can gather kinetics, operating conditions and process routes from papers and patents, then build them into an Aspen model you can check against the sources.

Find out why a run fails. It checks what Aspen still needs before a run, explains failed runs and warnings, and fixes them. Fixes that worked are kept on your machine for next time.

Optimize the process. Nine algorithms behind one tool — NSGA-II, NSGA-III, PSO, DE/GDE3, gradient (SLSQP), grid search and three Bayesian methods (BO, CBO, AFM-CBO) — for single- or multi-objective problems with constraints, plus pinch analysis and heat-exchanger match suggestions for heat integration.

Look things up instead of guessing. A reference for every block, stream and setting (aspen-mcp/docs) that the agent consults before it changes a field.

Built for working with people

An LLM agent that operates software on its own is hard to trust: you cannot see what it changed or why, or stop a mistake before it reaches the model. Aspen Co-pilot closes that gap with the Aspen Workbench, a desktop window beside Aspen Plus.

• See what the AI does. The flowsheet, the results and every change the AI made, with its reason, update live. Undo any change, or go back to a restore point.
• Stay in control. In review mode the AI's changes wait for your approval; larger changes arrive as proposals that only you can run.
• Define the work yourself with block programs. Assemble parameter sweeps, calculations, tables, charts and optimization studies — your own variables, objectives and constraints — from drag-and-drop blocks, with no code. The AI can write a program for you to edit, or run yours.
• Work together. Click any stream or block and Ask AI about it. In learning mode the AI explains the process principle behind each change.

The workbench has English and Chinese interfaces, and it also runs without an AI client (%USERPROFILE%\.aspen-mcp\aspen-workbench.cmd: no network, no tokens).

Install

You need Windows, a licensed Aspen Plus, and uv (winget install --id=astral-sh.uv -e; it brings its own Python).

Claude Code:

claude plugin marketplace add nckugese/Aspen_Co-pilot
claude plugin install aspen-plus@aspen-mcp

Codex: codex plugin marketplace add nckugese/Aspen_Co-pilot, then enable Aspen Plus in /plugins and trust its hooks in /hooks.

Then restart the client. In the first session it offers to finish setup (about 200 MB: Python, the server and the optimizer libraries); say yes, and restart once more. /aspen-doctor checks the result.

The plugin pins one release, so nothing changes until you update it (claude plugin update aspen-plus@aspen-mcp); it tells you when a newer release exists.

Other MCP clients (Claude Desktop, Gemini CLI, Codex without the plugin), the step-by-step version with checks, and troubleshooting: INSTALL.md.

Privacy

Everything the server keeps stays on your machine, under ~/.aspen-mcp/: its log, the change journal, restore points (copies of your models) and your error history. Your models, results and conversation are never sent anywhere by this project. The only network access is:

• the plugin's update check, at most once a day (git ls-remote against this repository);
• the first pinch_analysis call, which downloads Python 3.14 and OpenPinch through uv;
• installing and updating, which fetch the server and its dependencies through uv.

What your AI client sends to its own provider is governed by that client.

Feedback

Questions and bug reports: GitHub Issues. Security reports: SECURITY.md.

License

MIT — see LICENSE. Copyright (c) 2026 NCKU GESE Lab.

If you use this software in academic work, please cite this repository.

Disclaimer

Aspen Plus® and AspenTech® are trademarks of Aspen Technology, Inc. This project is an independent, unofficial integration and is not affiliated with, endorsed by, or sponsored by Aspen Technology, Inc. It does not include or distribute Aspen Plus; it controls a separate, user-provided installation, and you need your own valid Aspen Plus license.