How it is built

You ──▶ AI model (an app, or your API key) ──▶ MCP ──▶ openPASO (src/server.py)
                                                            │
   ┌──────────┬──────────┬────────┬─────────┬────────┬──────┴───┬────────┬────────┐
FEniCSx    deal.II       4C    NGSolve   skfem    Kratos     DUNE    FEBio   SPARTA
(Python)    (C++)     (YAML)  (Python) (Python)  (JSON)   (Python)  (XML)  (native)

Part	Where	What it does
Server	src/server.py	the MCP server the AI app or run_agent.py talks to
Backends	src/backends/<solver>/	one package per solver: the physics it supports, input writers, known traps
Registry	src/core/registry.py	loads the backends as plugins and finds which are installed
Tools	src/tools/	the tools the model calls
Knowledge	data/, src/backends/*/	settings and traps read from the version you actually installed
Coupling	src/core/coupling_driver.py	runs two participants until they agree (coupling)
preCICE bridge	src/core/precice_config.py	the couple_precice tool
Command line agent	run_agent.py, langgraph_eval/	Option B
Install check	check_install.py	the key-free check