Install Matlab

Install MATLAB on Linux with MPM

Goal

Install a chosen MATLAB release and set of products on a Linux machine non-interactively, after first installing the OS-level library dependencies that release requires. Uses MPM (the MATLAB Package Manager), the same mechanism the official MathWorks devcontainer feature and container images use.

How it works (two stages)

1. OS dependencies. MATLAB needs ~50 shared libraries (libGL, GTK, GStreamer, fontconfig, …). The exact list depends on the release AND the distro. The authoritative lists live in the MathWorks container-images repo: matlab-deps/<release>/<os>/base-dependencies.txt. The script detects the distro, maps it to the right list, downloads it, and installs it with the native package manager (apt-get or dnf/microdnf/yum).
2. MPM install. Downloads mpm from mathworks.com and runs mpm install --release=<R> --destination=<dir> --products <list> [flags], then symlinks the launcher to /usr/local/bin/matlab.

Does it need root?

• Yes for a normal install. Installing system packages (step 1) and writing to /opt/matlab + /usr/local/bin requires root. Run with sudo.
• MPM itself does not require root. If the OS dependencies are already present (e.g. inside a mathworks/matlab-deps base image) and you point --destination at a user-writable directory, MATLAB can be installed without root. In that case skip this script's root guard and run mpm install directly, or run this script's dependency stage once as root with --deps-only and do the MPM step separately.

Usage

The bundled script is scripts/install-matlab.sh.

# Latest supported release (R2025b), MATLAB only, to /opt/matlab/R2025b:
sudo bash scripts/install-matlab.sh

# A specific release + products:
sudo bash scripts/install-matlab.sh \
     --release R2024b \
     --products "MATLAB Simulink Signal_Processing_Toolbox" \
     --destination /opt/matlab/R2024b

# See exactly what would run without changing anything:
bash scripts/install-matlab.sh --dry-run --release R2024b

# Only install the OS dependencies (e.g. to bake into an image layer):
sudo bash scripts/install-matlab.sh --deps-only --release R2024b

Flags

• --release <Rxxxx> — MATLAB release (default R2025b). Supported R2019a–R2026a. To confirm the newest installable release, check that matlab-deps/<release-lowercase>/<os>/base-dependencies.txt exists upstream and that mpm --version is recent enough to know the release.
• --products "<list>" — space-separated product names for MPM (default MATLAB). Product names use underscores for spaces, e.g. Signal_Processing_Toolbox. See the MPM product list docs.
• --destination <dir> — install location (default /opt/matlab/<release>).
• --doc — also install documentation/examples (R2022b and earlier).
• --no-gpu (default) / --gpu — skip/include GPU libraries with Parallel Computing Toolbox. --no-gpu is only honored on R2023a and later; on older releases the script warns and installs GPU libs anyway.
• --deps-only — install OS dependencies, then stop (no MPM).
• --dry-run — print every command instead of executing (still fetches the dependency list so you can see the real package set).

Environment overrides

RELEASE, PRODUCTS, DESTINATION, DOC, INSTALLGPU are read as defaults; command-line flags win.

Workflow for Claude

1. Confirm the target. Ask/derive: which release, which products, where to install, and whether this is a bare host, a container, or a Dockerfile.
2. Check the platform. This skill is Linux-only. If Windows/macOS, stop and point at the interactive installer or matlab docs. Confirm the distro is a Debian/Ubuntu or RHEL/Fedora family (the script exits on others).
3. Preview first. Run with --dry-run and show the user the resolved dependency list and the exact MPM command before making changes.
4. Install. Run with sudo. For a Dockerfile, translate the two stages into RUN lines (deps then MPM) rather than calling the script, or COPY the script in and run it — either is fine; keep the dependency fetch in the build.
5. Verify. After install, check matlab -batch "disp(version)" or ls <destination>/bin/matlab. Note that licensing is separate: MATLAB still needs a license (e.g. a network license manager via MLM_LICENSE_FILE, or online licensing at first launch) — installation does not license it.

Notes & gotchas

• Licensing is not part of installation. MPM installs bits; the user must license MATLAB separately (network license manager, or sign-in at launch).
• Product names: spaces become underscores (Deep_Learning_Toolbox). A wrong name makes MPM fail — check the MPM product list if unsure.
• Unsupported OS/release combos: if the dependency list URL 404s, that distro isn't supported for that release; the script errors out clearly.
• Debian 13 (trixie): dpdk-named packages are filtered out (unavailable there), matching the upstream feature's behavior. The Ubuntu 24.04 list also pins libboost-*1.74.0, which trixie does not ship — the script rewrites those to the 1.83.0 series. Without that rewrite the whole apt-get install aborts.
• MPM behind a TLS-inspecting proxy: MPM's bundled libcurl uses its own CA store and ignores CURL_CA_BUNDLE/SSL_CERT_FILE, so a MITM proxy makes it fail with a bare Error: Download failed. Check the network connection and retry. The real cause is only in /tmp/mathworks_<user>.log (CURLcode=60, self-signed certificate in certificate chain). If direct egress to mathworks.com is allowed, run MPM with the proxy vars unset: env -u HTTPS_PROXY -u HTTP_PROXY -u https_proxy -u http_proxy mpm install ...
• Always check /tmp/mathworks_<user>.log when MPM fails; its stdout errors are generic.
• A handful of deps (libmosquitto*, libucx0, libpsm2-2, libhidapi-libusb0, libboost-log*) are only needed for optional features (MQTT, RDMA fabrics, hardware I/O); MATLAB starts without them.
• The script prefers wget but falls back to curl for downloads.