Teaching AI agents to run real engineering solvers, not toy demos
A grab-bag of MCP servers and agent skills that let Codex, Cursor, and Claude Code drive Abaqus, Fluent, LAMMPS, and friends on your actual machine.
What it does
CAE Agent Hub is a collection of MCP servers, reusable agent skills, and browser-based result viewers aimed at engineering simulation. Instead of AI clients producing offline toy examples, they connect to your installed CAE applications and solvers — Abaqus/CAE, ANSYS Fluent and Workbench, AEDT/HFSS, Altair HyperWorks, LAMMPS, OVITO, CalculiX, Peridigm, and a FreeCAD → Elmer → ParaView pipeline — and run real jobs. The repo also ships a large set of Abaqus-specific skills covering everything from geometry and meshing to fatigue analysis and topology optimization.
The interesting bit
The architecture is a clean separation of concerns: MCP servers provide tool access to live solver sessions, Skills encode the modeling and validation workflow knowledge, and the solvers themselves do the actual math. Several workflows are explicitly “evidence-oriented” — they gate results on verification steps rather than trusting the agent’s word, which is a refreshingly skeptical stance for the agentic-simulation space.
Key highlights
- MCP servers for ten-plus CAE ecosystems, from commercial heavyweights (Abaqus, ANSYS, HyperWorks) to open-source tools (CalculiX, LAMMPS, Elmer)
- Deep Abaqus skill coverage: 20+ skills spanning geometry, materials, contact, dynamics, thermal, fatigue, and optimization workflows
- Browser-based viewer for inspecting lightweight
result_mesh.jsonresults without keeping the solver open - CalculiX MCP includes two-stage sizing optimization and modal analysis with viewer-renderable mode shapes
- Bilingual (English/Chinese) documentation throughout
Caveats
- The Peridigm MCP and its fracture skill are marked as prototypes with only “verified profiles” — treat as experimental
- Some historical
text-to-caenaming lingers for viewer compatibility, which hints at an ongoing rename - You obviously need the underlying CAE applications and licenses installed; this repo is the glue, not the solvers
Verdict
Worth a look if you’re an engineer already using AI coding clients and want them to touch your real simulation stack. If you don’t have Abaqus or Fluent licenses sitting around, most of this will remain theoretical.
Frequently asked
- What is Cai-aa/CAE-Agent-Hub?
- A grab-bag of MCP servers and agent skills that let Codex, Cursor, and Claude Code drive Abaqus, Fluent, LAMMPS, and friends on your actual machine.
- Is CAE-Agent-Hub open source?
- Yes — Cai-aa/CAE-Agent-Hub is open source, released under the MIT license.
- What language is CAE-Agent-Hub written in?
- Cai-aa/CAE-Agent-Hub is primarily written in Python.
- How popular is CAE-Agent-Hub?
- Cai-aa/CAE-Agent-Hub has 1.1k stars on GitHub.
- Where can I find CAE-Agent-Hub?
- Cai-aa/CAE-Agent-Hub is on GitHub at https://github.com/Cai-aa/CAE-Agent-Hub.