A CLI for ML experimentation.

Define pipelines as DAGs, track metrics step-by-step, and monitor runs from a single control plane — all from your terminal.

Install

Get the CLI in seconds.

$curl -fsSL https://ml-code.com/install.sh | bash

Quickstart

Up and running in five steps.

1

Set up AI tools

mlc uses Claude to build pipelines from natural language. Export your Anthropic API key before starting the CLI.

$export ANTHROPIC_API_KEY=sk-ant-...
2

Start the CLI

Run the mlc command to open the interactive REPL. It connects to the control plane and starts a local MCP server.

$mlc
3

Log in

Inside the REPL, run /login. A browser window opens to complete authentication. Your credentials are stored locally.

>/login
4

Create and run a pipeline

Describe your pipeline in plain English. mlc creates the DAG, adds blocks, and asks you to confirm before executing.

>Create a pipeline with two blocks: one that generates 1000 random numbers and one that computes the mean and logs it

mlc will show each block it creates and ask for confirmation before running.

5

Monitor in the dashboard

Open the web dashboard to track block-level run status, view metric charts step-by-step, and inspect logs in real time.

$open https://app.ml-code.com

Dashboard

Monitor runs in real time.

The web dashboard gives you a live view of every pipeline run — block statuses, metrics, and logs, all in one place.

mlcmnist_trainingrunning

DAGs

mnist_training
data_prep
eval_sweep
Runs
3f8a1c
a21e90
7c4b02
load_datadone
trainrunning
evaluatepending

Metrics

train_loss

0.083

Logs

12:04:01INFOepoch 10/50
12:04:02INFOloss=0.083
12:04:03WARNlr decay step

Features

Everything you need.

DAG pipelines

Define blocks of Python code with explicit dependencies. mlc resolves execution order automatically.

Metric tracking

Emit scalar metrics per step with ctx.report(). Compare runs side-by-side in the control plane.

Structured logs

Use ctx.log() to emit structured log lines. Filter by block or level in the control plane.