Cfx MCP Tools

PyCFX-MCP tools

Use this skill when you are calling the standalone ansys-cfx-mcp CFX MCP leaf directly. The CFX leaf intentionally exposes a small tool surface so generic agents do not have to choose among many low-level helpers.

Tool surface

Expect these tools by default:

• session_status
• connect
• disconnect
• cfx_workflow
• cfx_model_context
• run_code
• validate_code

Do not assume direct MCP access to low-level helper names such as list_named_objects, get_state, find_api, get_help, solver_status, or summarize_setup. Use cfx_model_context or cfx_workflow instead.

For exact action names and parameter shapes, load tool reference.

Operating pattern

1. Call session_status first to learn whether CFX-Pre, Solver, or CFD-Post is already connected.
2. Use connect or cfx_workflow(action="start_pre") to start or attach to CFX-Pre when setup/model work is needed. Prefer an explicit connect mode ("launch" to start locally, "attach" to join a running server) so intent is unambiguous. mode="auto" (the default) infers it from the parameters.
3. Use cfx_model_context for bounded model inspection. Request one slice at a time, such as a summary, named-object lookup, API help entry, or targeted context.
4. Use cfx_workflow for lifecycle and artifact steps: import mesh, write .def, start solver, wait for solver, get .res, or open CFD-Post.
5. Use validate_code for explicit CFX Python snippets that are not covered by the routed workflow actions.
6. Use run_code only for safe, intentional inspection or edits requested by the user. Prefer read-only inspection first.
7. Call disconnect when the user asks to close the session or when cleaning up a dedicated run.

Token discipline

Keep context calls narrow. cfx_model_context accepts max_items. Use small values such as 10 or 20 unless the user explicitly asks for more. Do not dump the full model tree into the chat.

Safety rules

• Never invent CFX object paths or command signatures. Use cfx_model_context(action="api_help"), cfx_model_context(action="find_api"), or targeted context first.
• Do not use paths or assumptions from other products for CFX.
• Do not treat cfx_workflow as a free-form macro executor. It only accepts supported lifecycle/artifact actions.
• For solver launch requests, prefer cfx_workflow(action="start_solver"), then cfx_workflow(action="wait_solver"), then cfx_workflow(action="get_results_file").
• For custom setup changes, validate explicit PyCFX-oriented code, and then run it only after the user intent is clear.

Common recipes

Start CFX-Pre and import a mesh:

1. session_status
2. cfx_workflow(action="start_pre", params={})
3. cfx_workflow(action="import_mesh", params={"path": "C:\\path\\mesh.gtm"})
4. cfx_model_context(action="summary", max_items=20)

Launch a solver from an existing .def file and find the result file:

1. cfx_workflow(action="start_solver", params={"input_file": "C:\\path\\case.def", "mode": "serial"})
2. cfx_workflow(action="wait_solver", params={"timeout": 3600})
3. cfx_workflow(action="get_results_file", params={})

Inspect a named object without dumping the model:

1. cfx_model_context(action="find_named_object", params={"query": "inlet"}, max_items=10)
2. cfx_model_context(action="targeted_context", params={"paths_to_check": ["FLOW: Flow Analysis 1"], "named_object_types": ["boundary"]}, max_items=20)