Anaconda CLI reaches general availability as a single tool for packages, deployment and agents

Why Anaconda built a single CLI

Anaconda announced the public beta of the CLI last May. At that stage, one command configured conda and pixi against package sources an organization trusts. Since then, Anaconda says, more of the workflow has moved outside the terminal where those sources were set up. Projects go to a platform for deployment, coding agents add dependencies, and teams publish packages to their own channels.

Each of those steps has its own configuration and sign-in, according to Anaconda, and the sources configured in the terminal do not carry over. The result is that packages can enter a project from sources the organization never approved. The GA release is meant to close that gap.

One change to note: the Anaconda CLI now asks users to sign in. Anaconda says this lets it configure channels for the user's plan and connect the agent to Anaconda. Commands that connect to Anaconda services require sign-in.

Deployment and coding agent support

For deployment, ana platform app deploy sends a locally running project to Anaconda Platform, where it runs as a service inside the organization's perimeter. Related commands include:

• ana platform configure, which connects the CLI to an Anaconda Platform organization using the setup string from the organization's Local Setup page.
• ana platform app, which lists running apps, checks logs and deletes apps no longer needed.
• ana platform perimeter, which shows and switches the perimeter, defining the code, data, models and compute available to the work.

These platform commands work on macOS and Linux with the standalone installer.

For coding agents, ana mcp setup connects an AI assistant to Anaconda's remote Model Context Protocol (MCP) server. Before the assistant picks a package, it can check package metadata, security information and the organization's channel policies. Business and Custom plans include full vulnerability details. Setup also installs the Anaconda Intelligence skill, a markdown file added to the agent's directory that guides package selection, security review and troubleshooting. Anaconda notes that a package being available on a channel is not the same as the organization approving it, and the skill helps the agent tell the two apart.

The setup wizard configures assistants including Claude Code, Codex, Cursor, Devin, Kilo Code, OpenCode and VS Code. The MCP service is read-only: it does not install packages or enforce policy on other tools the agent uses.

Publishing, plan-based setup and updates

Teams can create public channels, or private channels, which are now in open beta, on anaconda.org. Names follow the organization/channel format, such as my-org/my-channel. ana channel create makes a channel, ana channel upload publishes packages to it, and ana channel remove deletes it. For publishing on each tagged release, the Upload Package action for GitHub Actions does the same job from continuous integration.

The main-x channel is an authenticated Anaconda channel that extends main with more than 13,000 additional Anaconda-curated conda packages. Running ana feature enable main-x makes the CLI check the user's Anaconda subscription and add the right channel URLs for the plan to conda, or to pixi with the --pixi flag.

The CLI now checks for new releases and notifies the user, but does not upgrade itself. Users run ana self update when they choose. Each release pins the versions of the tools the CLI manages, and the update also moves eligible managed tools to those versions, so it does not pull in upstream changes from that week. On a new machine, ana tool download miniconda downloads the right Miniconda installer for the platform without running it.

What it means for users

According to Anaconda, deployments, coding agent suggestions and team channels now run through the same trusted package sources, and builders spend less time switching tools and looking up repository URLs. Organizations that set package policy can have that policy visible to connected coding agents when they choose a package. conda and pixi continue to handle packages and environments, and the CLI connects them to the rest of the workflow without replacing either.

An Anaconda account is needed to authenticate and use features that connect to Anaconda services. On Windows, installation uses a PowerShell command, and on macOS (Apple Silicon) and Linux it uses a curl script. Beta users can update in place with ana self update. Issues and feature requests go to GitHub.