awesome-ai-cae

The CAE tools an AI agent can actually call

Open-source simulation, CAD & meshing tools for agentic / LLM-driven engineering — driveable headless via MCP, Python, or CLI (no GUI-only tools). The only CAE list with a weekly agent-callability ranking. Ranked by callability, not stars.

110+ tools · 3 MCP servers · 2 AI-Native · machine-readable JSON / CSV · weekly-regenerated ranking

Scope: agent-callable CAE/CAD/CAM tools, plus a small set of adjacent Datasets & Learning Resources for context.

🚀 Quickstart · 🏆 Index · 📊 Methodology

한국어 · 中文 · 日本語 · Deutsch · Français · Español · Português

Contents

• Quickstart
• AI-Readiness Index
• How the Score Works
• Core Engine Readiness
• MCP Servers
• CFD — Computational Fluid Dynamics
• FEA — Finite Element Analysis
• SPH — Smoothed Particle Hydrodynamics
• DEM — Discrete Element Method
• Visualization & Post-processing
• CAD & Geometry
• Mesh Generation
• Differentiable Simulation
• AI/ML for Simulation
• Surrogate Models & PINNs
• Optimization
• Data Formats & I/O
• Datasets & Benchmarks
• Learning Resources
• Star History

Quickstart — give your agent a CAE tool

Three tools here ship a Model Context Protocol server, so an agent (Claude Desktop, Cursor, Cline…) drives them with zero glue code. Add one to your MCP client config:

{
  "mcpServers": {
    "viznoir": { "command": "uvx", "args": ["viznoir"] }
  }
}

Exact launch command lives in each server's README. No MCP? Every other tool is Python/CLI-scriptable — your agent calls it the same way you would.

back to top

AI-Readiness Index

The headline metric: tools ranked by agent-callability — MCP, Python API, CLI, maintenance — not stars. Auto-updated weekly by readiness-score.py. Full table: READINESS.md · machine-readable: data/readiness.json.

🟢 2 AI-Native · 🔵 64 Agent-Ready · 🟡 25 Scriptable · ⚪ 23 Experimental — across 114 ranked tools (updated 2026-07-13). ✅ = install + import execution-verified. Full ranking →

back to top

How the Score Works

The AI-Readiness Score (0–100) ranks tools by how directly an autonomous agent can drive them — callability over popularity.

The five base signals (MCP + Python + CLI + Maintained + Adoption) total 100; pip is an additive bonus, and the final score is capped at 100. So a tool can reach 100 several ways, but only MCP servers clear the AI-Native bar.

Grades:

Scores regenerate weekly from README.md via readiness-score.py — fully reproducible, no hand-tuning. Open a PR adding a tool and a bot scores it automatically.

Verified vs declared

Honest about what's checked.

Verified (objective, reproducible) — live GitHub stars/activity; PyPI availability; and an install + import smoke-test (verify_install.py → data/verified.json) that spins up an isolated uv venv, runs pip install + import, and records the result. Tools that pass are marked ✅ in the Index. Current run: 8/10 flagship tools pass; the 2 misses are recorded honestly — Gmsh needs a system GL lib, DeepXDE needs a chosen backend.

Declared (from the entry's tags) — MCP and CLI/API. We link the server/CLI; we don't yet replay an end-to-end agent call.

Roadmap (deepening the moat) — execution-verified MCP handshakes and headless runs, plus a per-tool agent-call transcript, so the score reflects tools an agent has actually driven, not just ones that expose an interface.

back to top

Core Engine Readiness

A hand-curated editorial deep-dive on 17 foundational solvers — capability columns (Python binding, headless, Docker, AI-native) reflect maintainer judgment, ⭐ is live. This complements the auto-generated Index above, which stays the single source of truth for scores. Only 2 engines have MCP integration today.