CLI commands

Every command supports --json for structured output and follows the exit-code contract (0 ok, 1 error, 2 ran but a domain check failed). Run mcmc <command> --help for the authoritative, current flags.

Run inference

Command	Description	Key flags
run [input]	full workflow: fit, diagnose, record a run (omit the input on a terminal to pick a model; --report opens the result)	--data, --draws, --warmup, --chains, --adapt-delta, --seed, --backend, --entry, -o/--out, --stream-out, --prior, --algorithm, --thin, --adtype, --evaluation-mode, --parallel, --refit, --daemon, --julia-version, --package, --store
fit <spec>	run MCMC inference, write a samples file	-o/--out, --julia-version, --versions, --package-versions, --keep-going, --daemon
predict <spec> <samples>	draw posterior-predictive samples	-o/--out, --julia-version, --verbose

run takes a .jl or .stan model file, a spec, or a DoodleBUGS graph; fit and predict accept specs for both the Julia backends and Stan. For Stan, predict re-runs the model’s generated quantities block over the posterior draws. See Run inference and Predict.

Inspect runs

Command	Description	Key flags
browse	explore runs and models interactively (also what bare mcmc opens on a terminal)	--store
runs	list and manage recorded runs (runs list, runs prune --keep <n>)	--store
show [ref]	show one run’s settings and artifacts	--store
diagnose [target]	convergence diagnostics for a samples file	--rhat-max, --ess-min, --hdi-prob, --max-divergences, --warmup, --stdin, --store
sbc <input>	simulation-based calibration check	--simulations, --rank-draws, --bins, --daemon, plus the run sampler flags
loo [target]	cross-validated model fit (PSIS-LOO, WAIC)	--store, --verbose
compare <targets...>	rank runs by out-of-sample fit	--store, --verbose
summary [target]	posterior summary statistics	--var, --warmup, --stdin, --store
samples [target]	export the raw draws in a portable format	--to, -o/--out, --warmup, --stdin, --store
plot [target]	diagnostic plots for a samples file	--kind, --format, --var, -o/--out, --width, --height, --ascii, --hdi-prob, --bins, --max-lag, --color-by, --warmup, --stdin, --store
export <what>	copy a run’s artifact (samples/spec/record/bundle) to a visible file	--run, -o/--out, --force, --store
report [ref]	open a run in the report web app (report status, report stop manage the store server)	--store, --app-url, --no-open, --no-serve

A target is a samples file (MCMCChains JSON or ArviZ InferenceData JSON) or a run ref (latest, @N, an id prefix); it defaults to the latest store run. See Browse interactively, Diagnose convergence, Plot, and The run store.

Start a project

Command	Description	Key flags
init [dir]	seed a directory with a runnable example model and data	--force, --json
sandbox	open a throwaway shell seeded with an example model	--strict, --keep, --delete, --keep-dir, --name
convert <graph>	DoodleBUGS graph to a model file plus a fit-able spec	-o/--out, --seed

sandbox is the one interactive command; everything else runs unattended. See Convert DoodleBUGS.

Toolchain

Command	Description	Key flags
setup	install an inference toolchain: Julia via juliaup by default, CmdStan with --engine stan	--engine, --stan-version, --dry-run, --verbose
doctor	report every engine’s toolchain (or one with --engine)	--engine
engines	list known inference engines
julia version <sub>	manage installed Julia versions: list, status, add, remove, default, update, gc	--default (add), --verbose
stan version <sub>	manage installed CmdStan versions: list, status, add, remove	--verbose (add)
daemon <sub>	manage persistent Julia workers: status, stop
update	update mcmc in place: replaces a binary install from the latest release, or runs npm for an npm install	--check, --force, --json
mcp	run as an MCP server so an assistant can drive mcmc
skill <sub>	install the agent skill (skill install, skill show)	--project, --force, --json

See Manage Julia.