Colab CLI

Colab CLI is an open-source command-line tool published by the googlecolab GitHub organization under the Apache License 2.0. It brings Google Colab's cloud compute capabilities — including GPU and TPU runtimes — directly into terminal workflows, enabling headless automation and AI agent integrations without requiring an open browser tab. The project is written in Python and supports Linux and macOS only; Windows is explicitly not supported at this time.

What It Is

Colab CLI is a terminal-first interface for Google Colab that lets developers provision, control, and automate cloud VM runtimes from the command line. Rather than interacting with Colab through a browser notebook, users can spin up accelerator-backed VMs, run local Python scripts or Jupyter notebooks against those VMs, transfer files, and tear down resources — all through a structured set of colab subcommands. The project also ships a companion Colab MCP Server for in-notebook, interactive agent-assisted coding workflows.

Core Capabilities

The CLI is organized into four functional areas:

• Session management: colab new, colab sessions, colab status, colab stop, and colab url handle the full lifecycle of VM runtimes, including CPU, GPU (T4, L4, G4, H100, A100), and TPU (v5e1, v6e1) variants.
• Code execution: colab exec runs local .py files, .ipynb notebooks, or piped stdin against the remote kernel. colab repl opens an interactive Python REPL; colab console connects a raw TTY shell via tmux.
• File operations: colab upload, colab download, colab ls, colab rm, and colab edit manage the remote VM filesystem using the Jupyter Contents API.
• Automation utilities: colab auth handles GCP credential forwarding, colab drivemount mounts Google Drive, and colab install uses uv (with pip fallback) for fast dependency management.

Ephemeral Job Runner and Keep-Alive

Two design choices stand out for automation use cases. The colab run command provisions a fresh VM, executes a local script with forwarded arguments, retrieves output files, and automatically tears down the runtime in a single command — making it suitable for CI/CD pipelines and scheduled ML training jobs. A built-in background daemon provides automatic keep-alive, preventing idle VM termination without requiring an open browser session. Scripts can also use a shebang line (#!/usr/bin/env -S colab run --gpu L4) to run directly on remote accelerators as executable files.

Setup Path

Installation is straightforward via uv tool install google-colab-cli (recommended) or pip install google-colab-cli. Session tokens and metadata are stored at ~/.config/colab-cli/sessions.json; authentication supports both OAuth2 and Application Default Credentials (ADC). The repository includes deep-dive documentation covering session management architecture, execution design, file management, authentication providers, and the ephemeral job runner design, as well as eleven demo walkthroughs.

Current Status

The repository was created in April 2026 and last pushed in June 2026, indicating active early development. It is hosted under the official googlecolab GitHub organization and licensed Apache-2.0. The README notes that platform support is currently limited to Linux and macOS, and links to a sibling project (Colab MCP Server) for users who prefer interactive, agent-assisted notebook workflows.