simulink-integrate-python-code skill

Python Code Block Reference

Library path: simulink/User-Defined Functions/Python Code (R2025a+)

The Python Code block is the way to run Python inside a Simulink model. When a user asks to run a Python function, call a Python class, or use a Python library each timestep, this block is the tool — they may not name it explicitly.

Use it even when a MATLAB Function block or MATLAB System object calling py.* (coder.extrinsic) would produce the same numbers: those cannot build in Rapid Accelerator or model reference Accelerator mode, and they hide the Python behind a MATLAB block. The Python Code block works in both modes.

It converts inputs to Python objects, builds a dict keyed by symbol name, calls exec() with that dict as namespace, then converts outputs back to Simulink types. The interpreter is shared with MATLAB py.* — same process, modules, and state.

• Environment setup — pyenv, venvs, sys.path, module reloading
• Examples — verified programmatic patterns
• PythonObject patterns — class integration and persistent state

When to Use

• Running a line or block of Python each timestep inside a Simulink model
• Calling a Python function, class, or third-party library (numpy, scipy, …) from a model
• Persisting a Python object or numeric state across timesteps
• Passing a tunable parameter or a vector/matrix signal into Python
• The user describes the goal ("run my Python code in the model") without naming the block

When NOT to Use

• Pure MATLAB-side Python calls with no model involved — use py.* directly, no block needed
• MATLAB Function blocks or S-Functions — those are different mechanisms
• Running a pretrained deep-learning model (PyTorch/TensorFlow/ONNX) — use the Deep Learning Toolbox Model Predict blocks (PyTorch/TensorFlow/ONNX/Custom Python Model Predict), which run via Python co-execution and are configured through the block's model-file tabs, not SymbolSpec
• General Simulink model construction unrelated to Python — use the building-simulink-models skill

Workflow

Build the block in this order — ports do not exist until symbols are defined, so wiring comes last. The build takes five tool calls; batch the operations in each call:

1. Create and save the model (evaluate_matlab_code) — model_edit needs an existing model file.
2. Add and configure all blocks — one model_edit call. Put the code parameters (OutputCode, InitializeConditionsCode, TerminateCode, SampleTime) in add_block params. The Python Code block is created with no ports. The response maps each ref to its blk_N ID:

   [{"op":"add_block","type":"simulink/Sources/Constant","ref":"src","params":{"Value":"3"}},{"op":"add_block","type":"simulink/User-Defined Functions/Python Code","ref":"py","name":"Py","params":{"OutputCode":"y = u * 2.0","SampleTime":"1"}},{"op":"add_block","type":"simulink/Sinks/Out1","ref":"out"}]
   

3. Define symbols — one evaluate_matlab_code call; this creates the ports. SymbolSpec is a read-only handle reached only through get_param(blk,'SymbolSpec'), and model_edit cannot configure symbols — this is the one step that uses get_param directly. Call ss.addSymbol(...) and set each symbol's Scope/Type/Size in place.
4. Wire — one model_edit call using the returned IDs: [{"op":"connect","target":"blk_1.y1 -> blk_2.u1"},{"op":"connect","target":"blk_2.y1 -> blk_3.u1"}]
5. Simulate and read the result — one evaluate_matlab_code call. The simulation compiles the block, so it is the verification; skip separate inspection calls unless it fails.

The examples give the symbol definitions and code parameters for common patterns; read them only when the pattern isn't covered here.

To run the model programmatically and inspect results (sim/parsim, Simulink.SimulationInput, logsout), use the simulating-simulink-models skill.

Block Parameters

Parameter	Type	Notes
InitializeConditionsCode	string	Runs once at simulation start
OutputCode	string	Runs each timestep
TerminateCode	string	Runs once at simulation end
SampleTime	string	Default "-1" (inherited)
SymbolSpec	handle	Read-only param — get and mutate in place, never set_param back

What survives from InitializeConditionsCode to OutputCode: only imported modules (import m, import m as x, import pkg.sub) and Persistent symbols. Everything else in the init namespace is discarded — from m import f bindings, functions, classes, and plain variables (k = 3) raise NameError in OutputCode. Use the module-qualified form (import m → m.f(...)), repeat the from import in OutputCode, or store objects in a Persistent symbol.

SymbolSpec API

Class: Simulink.PythonFunctionBlock.SymbolSpec (handle object).

ss = get_param(blk, 'SymbolSpec');
sym = ss.addSymbol('name');   % Returns Symbol — then set properties
sym.Scope = 'Input';
sym.Type  = 'double';
sym.Size  = '1';
got = ss.getSymbol('name');   % Retrieve existing
ss.deleteSymbol('name');      % Remove

Handle object: mutations apply immediately to the block.

Symbol Properties

Property	Values	Notes
Name	char	Variable name in Python namespace
Scope	'Input' 'Output' 'Parameter' 'Persistent'
Type	'double' 'single' 'int8' 'int16' 'int32' 'int64' 'uint8' 'uint16' 'uint32' 'uint64' 'boolean' 'string' 'Bus: <BusObject>' 'fixedpoint' 'PythonObject'	PythonObject only for Persistent. Use lowercase 'boolean' — 'Boolean' is accepted by the setter but fails at compile ("not in scope"). A bus symbol names a Simulink.Bus object in the base workspace
Size	char	'1', '3', '[2 3]', '-1' (inherit)
Label	char	Port/mask display name
PortNumber	int	Port ordering

Data Type Mapping

Simulink	Python
scalar double	float
vector/matrix double	numpy.ndarray if numpy is available, else array.array. Shape follows Size ('[2 3]' ↔ (2, 3)); element (i,j) in MATLAB is [i-1, j-1] in Python — no transpose
string	str (output str becomes a MATLAB string). A root Inport fed from the workspace needs Interpolate off
bus (input)	dict keyed by element name — u["a"]
scalar integer	int
vector integer	numpy.ndarray if numpy is available, else array.array
PythonObject (Persistent)	any Python object

Limitations

• No variable-size signals or array of buses
• No continuous sample time with persistent variables
• No Dataflow Subsystems
• No rapid accelerator in For Each Subsystems
• No backward stepping with persistent variables
• Code generation supported (Python C API) — requires dev headers + interpreter linkage

---

Copyright 2026 The MathWorks, Inc.

---