Installation | MCMC.js

MCMC.js installs the mcmc command either as a single-file binary or from npm. Inference runs in a backend runtime (Julia, or CmdStan for Stan), which the CLI can install for you.

1. Install the CLI

Without Node.js

The install script downloads one self-contained binary for your platform (macOS and Linux, x64 and arm64) and needs neither Node.js nor npm.

curl -fsSL https://mcmcjs.github.io/install.sh | sh

It lands in ~/.local/bin by default; set MCMC_INSTALL_DIR to choose somewhere else, or MCMC_VERSION to pin a release. The script verifies the download against the release checksums and tells you if ~/.local/bin is not on your PATH. Upgrade by running it again.

With npm

You need Node.js 22 or newer.

npm install -g mcmcjs

Either way, check it:

mcmc --version

Installing both ways leaves two copies of mcmc, and whichever comes first on your PATH wins. The install script says so when it finds an existing copy, and mcmc doctor names the one that is running and warns about the other.

If your shell still runs an older copy after installing, it has cached the old location: run hash -r, or open a new shell.

Update

mcmc update           # update this copy in place
mcmc update --check   # only report whether a newer release exists

A binary install replaces itself from the latest release, verifying the checksum first; an npm install is updated with npm. --force reinstalls the current version, which repairs a damaged install.

Uninstall

curl -fsSL https://mcmcjs.github.io/uninstall.sh | sh

That stops the report server and removes the binary, keeping the cached Julia driver in case you reinstall; pass --all to remove that and the report server’s state too. Your runs are never touched, since they live in each project’s .mcmc directory, and neither are Julia or CmdStan. A copy installed with npm is reported rather than removed: take that one out with npm rm -g mcmcjs.

2. Provision Julia

Inference needs Julia and a managed set of packages (Turing.jl, JuliaBUGS, and friends). mcmc setup installs the toolchain through juliaup and precompiles the managed project, so you do not configure Julia by hand.

mcmc setup

Pass --dry-run to see what would be installed without making changes, or --verbose to see the full install and precompile output instead of a collapsed spinner.

The first setup downloads and precompiles a Julia project, which takes a few minutes. After that, fits start fast.

3. Verify

mcmc doctor reports every engine’s toolchain and tells you whether you are ready to fit.

mcmc doctor

Julia
juliaup  1.20.7  /home/you/.juliaup/bin/juliaup
julia    1.12.6  /home/you/.juliaup/bin/julia
ready for inference

Stan (CmdStan)
cmdstan  2.39.0  /home/you/.cmdstan/cmdstan-2.39.0
stanc    2.39.0  /home/you/.cmdstan/cmdstan-2.39.0/bin/stanc
make     4.4.1  make
c++      16.1.1  g++
ready for inference

The exit code is 0 as long as at least one engine is ready. Add --json for a machine-readable report, or --engine <id> to check one engine in the original flat format.

Stan (optional)

Stan models run through a local CmdStan, which mcmc setup can download and build for you.

mcmc setup --engine stan

Building CmdStan uses make and a C++ compiler (g++ or clang++) from your system; Linux and macOS are supported today. Pass --stan-version <v> to install a specific CmdStan release instead of the pinned default. Check readiness the same way as for Julia:

mcmc doctor --engine stan

If you already have CmdStan, the CLI also finds it through the MCMCJS_CMDSTAN or CMDSTAN environment variable, or at ~/.cmdstan, shared with other Stan interfaces.

Libraries

The CLI is the unscoped mcmcjs package. The internal libraries it is built from publish under the @mcmcjs/* scope (for example @mcmcjs/core and @mcmcjs/diagnostics). You do not need to install those directly to use the CLI; see Packages if you want to build on them.

Next steps

Fit and diagnose your first model in the Quickstart.