CFAST MCP - Simulador de Incêndios com IA
bewygs/cfast-mcp
Simule incêndios em ambientes pelo Claude com CFAST e PyCFAST
Ver no GitHubServidor MCP que permite a um assistente de IA construir, executar e analisar simulações de incêndio CFAST (modelo NIST) através de conversação natural. Expõe ferramentas para criar compartimentos, adicionar materiais, ventilações, focos de incêndio e dispositivos, executar simulações e extrair resultados.
Instalação
npx convez-ai bewygs/cfast-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Simulação de dinâmica de incêndio em edifícios e ambientes fechados
- Análise de propagação de fumaça e temperatura para estudos de segurança
- Prototipagem rápida de cenários de incêndio sem interface gráfica
Veredito

Ferramenta especializada e bem estruturada para simulação de incêndios via IA. Útil para engenheiros de segurança, pesquisadores e arquitetos que precisam modelar cenários de incêndio conversacionalmente, mas com nicho técnico bastante específico.
README
A tradução para o português deste README ainda está na fila.
CFAST MCP
CFAST MCP is an MCP server that lets an AI assistant build, run, and analyze CFAST (Consolidated Fire and Smoke Transport, NIST) fire simulations through conversation. It is built on top of PyCFAST and exposes the CFAST model as a set of tools. The AI assistant is able to create a model, add compartments, materials, vents, fires and devices step by step, run CFAST, and make summaries of the results.
Example
Ask your assistant something like:
Create a 4 m × 3 m × 2.5 m room with a door (0.9 × 2 m) to the outside and a fire growing to 1 MW in 300 s. Run it and give me the peak upper-layer temperature then show me the folder where you create the file, so I can inspect it.
Results will probably look like this:
Tools
| Group | Tools |
|---|---|
| Create & configure | create_model, update_simulation |
| Components | add_* / update_* for materials, compartments, wall vents, ceiling/floor vents, mechanical vents, fires, devices (targets & detectors), surface connections |
| Inspect | inspect_model (summary, optional .in file), get_model_files |
| Run & results | run_model, get_results (bounded previews and per-column min/max/final stats) |
Results are returned to the AI assistant as small text summaries. The generated files (.in, output .csv, logs) are written in a temporary directory while the session is active. Use get_model_files to locate them if you want to open them directly.
Note: models live in memory for the lifetime of the server process. Restarting the server (or your MCP client) will delete them.
Installation
Requires Python 3.10+ and CFAST 7.7.0+.
uvx (Recommended)
Install uv, then add cfast-mcp directly in your client configuration:
{
"mcpServers": {
"cfast": {
"command": "uvx",
"args": ["cfast-mcp"],
"env": { "CFAST": "/path/to/your/cfast/executable" }
}
}
}
Claude Code
If you use Claude Code, a single command registers the server:
claude mcp add cfast -e CFAST=/path/to/your/cfast/executable -- cfast-mcp
Pip
Create a virtual environment and install from PyPI:
python -m venv venv
source venv/bin/activate # Linux/macOS
venv\Scripts\activate # Windows
pip install cfast-mcp
Then add cfast-mcp to your client configuration:
{
"mcpServers": {
"cfast": {
"command": "cfast-mcp",
"env": { "CFAST": "/path/to/your/cfast/executable" }
}
}
}
CFAST Installation
Download and install CFAST from the NIST CFAST website or the CFAST GitHub repository. Follow the installation instructions for your operating system and ensure cfast is available in your PATH. If CFAST is installed in a non-standard location, you can manually specify the path by setting the CFAST environment variable to point to the CFAST executable.
export CFAST="/path/to/your/cfast/executable" # Linux/macOS
set CFAST="C:\path\to\cfast.exe" # Windows (cmd)
$env:CFAST="C:\path\to\cfast.exe" # Windows (PowerShell)
Development
git clone https://github.com/bewygs/cfast-mcp.git
cd cfast-mcp
uv sync --extra dev # install dev dependencies
uv run pytest # run tests
uv run ruff check --fix . # lint
uv run mypy src/ # type-check