AI news, problem solutions & tool picks
DigestX

Tool Suggestions

Claude Code Cloud Now Sets Up Your Repo Environment by Itself

· 5 min read

A laptop displaying source code on two screens
Photo via Pexels

Claude Code cloud can now prepare a repository's environment on its own. Anthropic's developer account, @ClaudeDevs, announced the change on November 6, 2026, for subscribers on the Pro, Max and Ultra plans: Claude works out which dependencies, network access and API keys a project needs, saves the configuration for later sessions, and keeps the keys out of Claude's reach.

“We've made it easier to get your repo running in Claude Code cloud sessions on Pro, Max, and Ultra,” the post reads. “Claude figures out what dependencies, network access, and API keys it needs, then saves the environment for next time. API keys are saved securely, so Claude never sees them.”

The problem it solves

A repository that runs without trouble on a laptop often stalls on a clean machine, whether a library is missing, a host is blocked by the network, or the key for an outside service is nowhere to be found. Until now it fell to the user to discover the problem, fix it and try again, in a loop that could eat the first half hour of a session.

A cloud session, the version of Claude Code that runs on Anthropic's virtual machines rather than on the user's own computer, keeps going even with the laptop closed. Cloud sessions start from claude.ai/code, the desktop app, the mobile app or the terminal with claude --cloud, normally with a GitHub repository connected. The announcement promises to cut out the wasted setup loop at the start of that session.

What a cloud environment contains

Each cloud session runs in a cloud environment, and the official documentation describes four things that environment can hold. Network access decides which domains the session can reach, with four levels — None, Trusted, Full and Custom. Trusted is the default and allows package registries and a list of approved domains, while Custom lets you set your own list of domains. Environment variables are written in .env format, one key=value pair per line, and any command Claude runs can read them. A setup script is a Bash file that runs before Claude starts and installs whatever is missing on the machine.

The fourth piece, network secrets, is the one that matters most here. Per the documentation, network secrets are API keys or tokens stored in the environment, which sessions use without seeing them. Anthropic's agent proxy adds the key to requests bound for the hosts the user lists, after each request has left the virtual machine, so the value stays outside the VM. Network secrets are available on the Pro, Max, Ultra, Team and Enterprise plans.

Before this announcement, a cloud session started from an environment called Default, which carries no configuration of its own: network access set to Trusted, no environment variables or setup scripts, and only the tools already installed on the machine. A project with particular needs meant opening the environment picker, creating a new environment and filling in the fields by hand. Now, per the post, Claude works out what is needed by itself, decides which dependencies to install, which network hosts to reach and which API keys it requires, then saves the environment as it came out. The next session on the same repository does not repeat the work.

What the announcement does not say

The announcement describes the property — keys saved securely, never visible to Claude — without naming the mechanism. The description matches that of network secrets, but the post does not say they are the same thing, and it does not say how the user hands over the value of a key that Claude has flagged as necessary.

The documentation page on cloud environments, read on November 7, still explains manual configuration and does not mention automatic detection, so it is unknown which files Claude reads to decide or by what rules. The post also does not say whether the user approves the network hosts Claude opens or the values of the keys it needs. Before entrusting the setup to a repository with production credentials, it is worth opening the saved environment and checking what it contains.

Two more limits are worth knowing. The announcement does not say automatic setup replaces the setup script, and the documentation still describes the script, the variables and the network secrets as fields of the environment. And the announcement names Pro, Max and Ultra. Cloud environments exist for Team and Enterprise too, but for those plans no official source confirms automatic setup, so anyone working inside an organization should check with their Owner.

Why the keys part matters most

An API key is the string that authorizes a program to use a service on the user's behalf — a database, a payments provider, an artificial intelligence model. In a cloud environment the most convenient shortcut would be pasting it among the variables, but the documentation warns that those values can be read by anyone who uses that environment. With a network secret the problem shifts: you can ask Claude to make a curl call to the API, which responds as if the key were there, while the key shows up neither among the variables nor in a file.

The principle has a concrete reason behind it. A SOCKS5 flaw in Claude Code's network sandbox stayed open from October 20, 2025 to April 1, 2026 and, according to researcher Aonan Guan, who documented it, allowed credentials and source code touched by the agent to be smuggled out. A key that never enters the virtual machine cannot be lifted from there. That is why it matters whether automatic setup actually goes down this road — and why it is worth verifying, rather than assuming, once the feature reaches your account.

Where it fits in Claude Code's cloud push

The news lands after seven busy weeks for cloud sessions. On September 17, Anthropic introduced Claude Code Projects and parallel threads in the cloud. On September 23 it took cloud sessions out of research preview, with a one-time credit to try them. October 16 brought gzip on the conversation and a faster start, and in early November came message threads, which make Claude Code respond 35% sooner. The common thread is the time between the command and the first useful work: first the speed of the start, now the preparation of the environment.

What remains to be seen is how often Claude gets it right on the first try with a real repository — say a monorepo, a private registry and three outside services. A cloud that is more and more ready to use, with one question still open on the keys.