CAE Agent Hub

Connect AI to real CAE. Keep the evidence behind every result. English · 简体中文

A hub of MCP tools, reusable skills, and result viewers for engineering simulation. Connect clients such as Codex, Claude Code, Claude Desktop, and Cursor to modeling, meshing, solving, postprocessing, and validation workflows.

Simulation gallery · Tool index · Quick start · Result viewer

Also includes Subagent resources and Agentic PD workflows with curated recorded demos.

Representative simulation results

One representative case per software module with verifiable results. Modules without a suitable result are omitted for now. Click an image for full size; see the full atlas for additional cases, checks, and provenance.

Case notes document image types, units, display scales, and validation scope. Engineering use requires the relevant engineer’s review.

Tools and workflow navigation

Choose an adapter by application. Each README covers installation, client configuration, and limitations. This is a source index; runtime availability depends on local software, versions, licenses, and bridge state.

Skills: Full index · Abaqus workflows · Ansys structural workflow · Cylinder O-grid

Quick start

Clone

git clone https://github.com/Cai-aa/CAE-Agent-Hub.git
Set-Location .\CAE-Agent-Hub

If you cloned the repository before the rename, the old text-to-cae remote may still work through GitHub redirects. Updating the remote URL is clearer:

git remote set-url origin https://github.com/Cai-aa/CAE-Agent-Hub.git

Use an MCP server

Each adapter has its own dependencies and launch command; open its README from the table above. Install the required CAE application and confirm its license, then follow that adapter’s environment, dependency, and MCP registration steps. Install the in-application bridge where required.

Start with environment detection or a read-only model query, then run an example. Use a separate output directory and retain inputs, solver logs, and validation records.

Use skills

Skills are instruction modules, not solver binaries. Copy the complete skill directory into your agent’s skill directory or attach the relevant SKILL.md as project context.

Codex example:

New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.codex\skills" | Out-Null
Copy-Item -Recurse -Force ".\Skill\abaqus\analysis\abaqus-static-analysis" "$env:USERPROFILE\.codex\skills\abaqus-static-analysis"

Suggested prompt:

Use the abaqus-static-analysis, abaqus-mesh, abaqus-job, and abaqus-odb skills. Check the Abaqus environment and inspect the current session first. Build a cantilever static-analysis case in a separate directory, mesh and solve it, extract ODB results, check force balance, and report the actual files, commands, and incomplete checks.

Text to CAE Viewer

The viewer is still included as the browser result-inspection layer. It can display cases that contain result_mesh.json even when the solver is not installed.

Set-Location .\viewer
npm.cmd install
npm.cmd run dev

Open the Vite URL, usually:

http://127.0.0.1:4178/

Agentic PD recorded demos

Run the viewer as above, then open the Peridigm entry at /peridigm.html. Choose either demo:

• /peridigm.html?demo=elastic — two-point elastic verification: 6 actual frames, source gold and an independent oscillator comparison.
• /peridigm.html?demo=fracture — precracked tensile damage demonstration: 3,200 material points, 31 actual display frames and independently reconstructed bond failure.

Replay requires neither Peridigm nor SSH. npm run build bundles both datasets for static hosting. These are recorded verified results, not a browser solver or a connected AI service. New solves use the Peridigm CLI/MCP and its Skill. See demo provenance and limits.

Repository Layout

CAE-Agent-Hub/
  MCP/
    Abaqus/
    CalculiX/
    Adams/
    Peridigm/
    COMSOL/
    CST/
    FEP-Agent-Hub/
    HyperWorks/
    LAMMPS/
    OVITO/
    Ansys/
      AEDT MCP/
      Fluent MCP/
      Workbench MCP/
      PyAnsys/
  Skill/
    abaqus/
      core/
      modeling/
      setup/
      analysis/
      execution/
      postprocessing/
      optimization/
      reference/
  Subagent/
  models/
  viewer/
  assets/readme/   # Curated figures and provenance
  docs/           # Case notes and training material

Source Control Policy

The repository should include reusable source, documentation, tests, examples, templates, and skill instructions. Curated public replay fixtures under models/peridigm-* are an explicit exception to excluding generated run data: they include only selected fields, verified summaries, canonical inputs and hashes, with attribution and capability limits. Full native jobs and private runtime state stay outside the repository.

It should not include:

• CAE software binaries or licenses.
• Private machine paths or credentials.
• Virtual environments and package caches.
• Raw solver outputs such as ODB, case/data, AEDT results, Workbench projects, and runtime logs.
• Large local result artifacts that can be regenerated.

Curated documentation images may live in assets/readme/ with provenance, case definitions, and scope notes.

Build

Build the frontend viewer:

Set-Location .\viewer
npm.cmd run build

Roadmap

The hub is designed to grow toward more mainstream simulation ecosystems:

• More solver-specific MCP servers.
• More productized skill packs for repeatable modeling and validation workflows.
• Shared result export formats for viewer and report generation.
• Safer install prompts and verification scripts for each CAE application.

Contributing a New MCP or Skill

When extending the hub:

1. Keep reusable source code and workflow instructions together.
2. Provide examples and tests where appropriate.
3. Provide a README for the new component.
4. Keep private paths, credentials, licenses, virtual environments, and generated solver results out of source control.
5. Follow the existing repository structure and naming conventions.
6. For Skills, preserve the complete SKILL.md-based workflow and supporting references/assets.

License

This repository is released under the MIT License. See LICENSE for details.

Peridigm development module

The Peridigm module adds a shared CLI/MCP workflow, confirmed specifications, bounded repair and material-point viewer export. Qualified profiles cover two-point elastic code verification and a synthetic precracked tensile damage demonstration. Engineering strength and experimental accuracy remain unverified. See its architecture map and Skill.