Call Python from MATLAB

Call Python libraries from MATLAB using modern APIs, correct environment setup, and structured error recovery.

When to Use

• Writing MATLAB code that calls Python via py.*, pyrun, or pyrunfile
• MATLAB code fails with ModuleNotFoundError, Python is not configured, or other Python-related errors
• User asks to set up Python for MATLAB or create a virtual environment
• User asks to install a Python package for use in MATLAB

When NOT to Use

• Calling MATLAB from Python (MATLAB Engine API for Python — different direction)
• Building MEX files or Python C extensions
• Pure Python development with no MATLAB involvement
• MATLAB Compiler or MATLAB Production Server deployment

Rules

NEVER use pyargs

Use MATLAB name=value syntax for Python keyword arguments. pyargs is outdated (name=value available since R2021a).

% TEMPLATE — not executable (requires scikit-learn)
rfc = py.sklearn.ensemble.RandomForestClassifier(n_estimators=int32(100), random_state=int32(42));

NEVER use pyversion

pyversion is not recommended. Use pyenv, which supports terminate, ExecutionMode, and environment switching.

ALWAYS use the Bash tool for pip installs and venv creation

Do not use MATLAB system(). The Bash tool provides streaming output and avoids complex string escaping.

ALWAYS follow recovery procedure before any pip install or venv creation

BEFORE running pip install or creating virtual environments, follow references/environment-setup.md. Do NOT infer Python paths from error tracebacks — those paths show importlib internals, not the correct install target.

ALWAYS ask user confirmation before calling terminate(pyenv)

terminate(pyenv) restarts the Python process and clears all Python state (imported modules, variables, open connections). Always inform the user what will be lost and get explicit consent before calling terminate.

ALWAYS wrap with integer types when Python expects int

MATLAB passes double by default. Python functions expecting int receive float, causing TypeError. Use int32() for counts, indices, and size arguments. If the value won't fit in 32 bits, use int64() or uint64() to prevent overflow.

ALWAYS use pystringarray for string arrays (R2026a or later)

When passing MATLAB string arrays to Python in R2026a or later, use pystringarray. Requires NumPy 2.0 or later. In R2025b and earlier, or when NumPy < 2.0, use py.list(cellstr(...)).

% R2026a and later:
names = ["Alice", "Bob", "Charlie"];
pyNames = pystringarray(names);

% R2025b and earlier:
names = ["Alice", "Bob", "Charlie"];
pyNames = py.list(cellstr(names));

Workflow

When calling Python from MATLAB, follow this workflow.

1. Write the Python code using modern syntax

result = py.module.function(arg1, arg2, kwarg1="value", kwarg2=42);

Prefer py.module.function(...) for direct function calls — it's readable, integrates with the MATLAB workspace, and supports tab completion. Reserve pyrun for multi-statement scripts where variables and imports persist in Python memory across calls (REPL-like execution). Reserve pyrunfile for running a standalone .py file (equivalent to running from the command line; no state carries over).

2. Run the code

Execute via mcp__matlab__evaluate_matlab_code (for inline code) or mcp__matlab__run_matlab_file (for script files).

3. If it succeeds — done

4. If it fails with a Python error — triage

Follow references/environment-setup.md ONLY for environment-related errors:

1. ModuleNotFoundError
2. MATLAB error about unresolved py.* name (e.g., "Unable to resolve the name 'py.numpy'")
3. pyenv().Executable is empty
4. py or pyenv calls error stating environment is not supported
5. Any other error suggesting a missing library or unconfigured environment

For all other Python errors (ValueError, TypeError, logic errors in the Python code itself) — debug the code directly. These are code bugs, not environment issues.

Key Functions

Function	Purpose	Available From
pyenv	Query/configure Python environment	R2019b
pyrun	Execute Python statements; variables persist across calls	R2021b
pyrunfile	Execute a standalone .py file	R2021b
pystringarray	Pass MATLAB string array to Python	R2026a

InProcess vs OutOfProcess

• OutOfProcess: Launches Python in a separate process. Mitigates third-party library conflicts between MATLAB and Python, therefore works with most Python packages without conflict. terminate(pyenv) restarts Python without restarting MATLAB. Can switch environments freely.
• InProcess: (default) Launches Python within the same MATLAB process. PREFER for better performance. Cannot terminate. If Python is already loaded in InProcess mode, the user must restart MATLAB entirely to change Python configuration.

In ExecutionMode="OutOfProcess", when working on a custom Python module and you need to make changes after loading the module, or when you install packages mid-session, ALWAYS call terminate(pyenv) before calling the module again.

In ExecutionMode="InProcess", when working on a custom Python module and you need to make changes after loading the module, ALWAYS reload the module by running:

clear all;
clear classes;
mod = py.importlib.import_module("<module_name>");
py.importlib.reload(mod);

Patterns

Calling a Python function with keyword arguments

rfc = py.sklearn.ensemble.RandomForestClassifier(n_estimators=int32(100), random_state=int32(42));
rfc.fit(XTrain, yTrain);
predictions = double(rfc.predict(XTest));

Configuring MATLAB to use a virtual environment

After creating a virtual environment (venv) using the Bash tool, point MATLAB at it:

% Terminate current Python if loaded
terminate(pyenv);

% Point at venv executable
pyenv(Version="/path/to/.venv/Scripts/python.exe");

% Verify
disp(pyenv().Executable);
py.numpy.array([1, 2, 3]);

Conventions

• When a keyword argument name conflicts with a MATLAB variable, append Val (e.g., py.complex(imag=imagVal) not py.complex(imag=imag))
• AVOID "activating" a virtual environment — MATLAB doesn't use activation scripts. Point pyenv(Version=...) at the venv executable directly.

---

Copyright 2026 The MathWorks, Inc.

---