English | 한국어

parapilot

Headless CAE post-processing MCP server for AI coding assistants.

Landing Page · PyPI · Issues

pip install mcp-server-parapilot

8.8M cell SA-DDES automotive CFD — single MCP tool call, no GUI

What it does

parapilot lets AI assistants (Claude Code, Cursor, Gemini CLI) render CFD/FEA simulation results without a GUI. It talks to ParaView headless or uses VTK directly to produce PNG screenshots, statistics, and animations from OpenFOAM, VTK, CGNS, and 30+ other formats.

Quick Start

Claude Code (plugin)

claude install kimimgo/parapilot

Then in a conversation:

"Render the pressure field from cavity/cavity.foam with a jet colormap"

Standalone MCP server

{
  "mcpServers": {
    "parapilot": {
      "command": "mcp-server-parapilot"
    }
  }
}

Docker (GPU headless)

docker compose up -d

Requires NVIDIA Container Toolkit for GPU rendering.

Docker (CPU-only, no GPU)

docker compose up parapilot-cpu -d

Uses OSMesa software rendering — works on any machine without a GPU.

Examples

See examples/ for complete workflow pipelines:

• thermal_analysis.json — Conjugate heat transfer post-processing (8 steps)

Tools (18)

Tool	Description
inspect_data	File metadata — fields, timesteps, bounds
render	Single-field PNG screenshot
slice	Cut-plane visualization
contour	Iso-surface visualization
clip	Clipped region visualization
streamlines	Vector field flow visualization
cinematic_render	Publication-quality render (SSAO, PBR, 3-point lighting)
compare	Side-by-side or diff comparison of two datasets
probe_timeseries	Sample field at a point across timesteps
batch_render	Render multiple fields in one call
preview_3d	Export to glTF/glB for interactive 3D viewing
extract_stats	Min/max/mean/std for fields
plot_over_line	Sample values along a line
integrate_surface	Force/flux integration over surfaces
animate	Time series or camera orbit animation
split_animate	Multi-pane synchronized animation (GIF)
execute_pipeline	Full pipeline DSL for advanced workflows
pv_isosurface	DualSPHysics bi4 → VTK surface mesh

Resources (11)

URI	Content
parapilot://formats	Supported file formats and readers
parapilot://filters	Available filter parameters
parapilot://colormaps	Colormap presets
parapilot://cameras	Camera angle presets + auto-camera PCA
parapilot://cinematic	Lighting, materials, backgrounds, quality presets
parapilot://representations	Render representations
parapilot://case-presets	Domain-specific case presets
parapilot://physics-defaults	Physics-aware rendering defaults
parapilot://pipelines/cfd	CFD pipeline examples
parapilot://pipelines/fea	FEA pipeline examples
parapilot://pipelines/split-animate	Split animation examples

Showcase

All renders below are generated by single MCP tool calls — no post-processing.

Bone isosurface (contour)	Carotid blood flow (streamlines)	Stanford Dragon (render)
Volume ray-casting (render)	HVAC airflow (streamlines)	Mesh clip (clip)

12 engineering domains — Marine, Mechanical, Aero, Electronics, Biomedical, Geoscience, Environmental, Chemical, Structural, Naval, Scientific, Urban.

See the full gallery at kimimgo.github.io/parapilot.

Architecture

┌─────────────────────────────────────────────┐
│  AI Assistant (Claude / Cursor / Gemini)    │
│  ↕ MCP protocol (stdio)                    │
├─────────────────────────────────────────────┤
│  parapilot MCP Server (FastMCP)             │
│  ├── tools/     18 MCP tools                │
│  ├── resources/ 11 MCP resources            │
│  └── prompts/   3 MCP prompts               │
├─────────────────────────────────────────────┤
│  Engine Layer (VTK direct API)              │
│  ├── readers    OpenFOAM, VTK, CGNS, ...    │
│  ├── filters    Slice, Contour, Clip, ...   │
│  ├── renderer   Off-screen VTK rendering    │
│  ├── camera     Preset + custom positions   │
│  ├── colormaps  Scientific color schemes    │
│  ├── overlay    Scalar bars, labels, text   │
│  ├── physics    Auto-detect field types     │
│  └── export     PNG, VTK, CSV output        │
├─────────────────────────────────────────────┤
│  Core Layer                                 │
│  ├── compiler   Pipeline → VTK script       │
│  ├── runner     Local / Docker execution    │
│  ├── registry   Filter & format schemas     │
│  └── output     Result collection           │
└─────────────────────────────────────────────┘

Workflow

inspect_data → render / slice / contour → extract_stats → animate

1. Inspect — discover fields, timesteps, bounds
2. Visualize — render, slice, contour, clip, streamlines
3. Extract — statistics, line plots, surface integrals
4. Animate — time series or multi-pane comparison

Supported Formats

OpenFOAM (.foam), VTK (.vti/.vtp/.vtu/.vtm), CGNS (.cgns), Ensight (.case), Exodus (.exo), STL (.stl), PLY (.ply), OBJ (.obj), and 30+ more via VTK readers.

Contributing

We welcome contributions! Check out our open issues — especially those labeled good first issue.

git clone https://github.com/kimimgo/parapilot
cd parapilot
pip install -e ".[dev]"
pytest                     # 748 tests
ruff check src/ tests/     # lint
mypy src/parapilot/        # type check

See CONTRIBUTING.md for detailed setup, architecture guide, and how to add new filters/readers.

vs Alternatives

	parapilot	LLNL/paraview_mcp	Kitware/vtk-mcp
Rendering	Headless VTK + ParaView	GUI-attached only	None (docs search)
Tests	748	0	0
Docker	GPU (EGL)	No	No
MCP Tools	18	23	3
Plugin	Claude Code plugin	No	No

Known Limitations

• Validated with VTK example datasets only — large-scale industrial data (100M+ cells) is unverified
• Headless rendering means no real-time visual feedback to catch LLM-generated parameter errors
• No simulation steering, multi-physics coupling, or uncertainty quantification
• When ParaView natively integrates MCP, this wrapper approach may become redundant

Contributors

kimimgo	NikanEidi

License

MIT