Starting a cloud coding session is about to get simpler for Claude Code users on paid plans. Anthropic says its cloud sessions can now work out what a repository needs to run — which dependencies to install, which network hosts to reach, and which API keys the work requires — and then save that setup for the next session. The change covers the Pro, Max and Ultra plans, and it removes the manual environment configuration that used to stand between opening a session and getting work done.
What changes
Until now, a cloud session started from a bare Default environment unless the user had configured one by hand. That default allows trusted network access to package registries and a short allowlist, but it defines no environment variables and runs no setup script, so sessions began with only the pre-installed tools.
Anyone whose project needed a database client, a private registry, or an external API had to wire all of that up first, before Claude wrote a line of code. The new behaviour hands that job to Claude: it inspects the repository, decides what the session needs, and stores the environment so the next session on the same repository starts ready.
Keys Claude never sees
The most delicate part of the change is the handling of API keys. Anthropic says keys are saved securely so Claude never sees them, and its documentation explains how: a stored key is held as a network secret, and the company’s agent proxy attaches it to requests for the hosts the user lists.
The proxy adds the key only after each request has left the session’s virtual machine, so the key itself stays outside the VM. It never appears in the session’s environment variables or in any file, and requests made by the setup script never receive a secret at all, because the proxy only attaches once Claude Code has launched.
The announcement names three plans — Pro, Max and Ultra — and says nothing about Team or Enterprise, so whether automatic setup reaches those plans is not yet established. Network secrets themselves are available more widely, on Team and Enterprise as well, but the new hands-off setup is described only for the three individual plans.
Cloud sessions start from the web, the desktop and mobile apps, or the terminal with a single command, and each one runs on Anthropic’s own machines, so the work keeps going with the laptop closed. That is the point of the cloud version: the session is no longer tied to the machine in front of the developer.
Our read
Setup friction was the quiet tax on cloud coding agents. The pitch of a cloud session is that the agent keeps working while the developer is away, yet every new repository began with a configuration chore that only a human could do — and that chore had to be repeated, or at least checked, for each project.
By letting Claude infer the environment from the repository itself, Anthropic removes the last manual step before the agent starts — the step most likely to put a casual user off trying cloud sessions at all. It is a small change on paper, but it attacks the moment when a user decides whether the cloud version is worth the effort.
The move also keeps Anthropic in step with its rivals, who have been racing to make their agents start with less hand-holding. OpenAI’s Codex runs cloud tasks from a saved machine state, and GitHub has been rebuilding its own plumbing so thousands of agents can push at once, while Cursor keeps adding ways to launch agents without setup.
None of those removes configuration entirely, and Anthropic’s answer — infer it, save it, reuse it — is the most direct version yet. The open question is how much a user should trust an agent to choose its network access and keys, and the announcement does not say which files Claude reads to decide, or whether users approve each setting.
Why It Matters
Cloud coding agents only earn their keep when starting one is nearly free. Every minute spent configuring an environment is a minute the agent is not working, and it is also a reason to stay on the local terminal instead.
Automatic setup lowers that barrier for the three paid plans where Anthropic is courting individual developers, and the key-handling design — secrets the agent can use but never see — answers the security objection before it is raised. If the saved environments prove reliable, the cloud session stops feeling like a second setup and starts feeling like the default.
Background
Claude Code is Anthropic’s coding agent, and its cloud sessions run on the company’s own machines rather than the developer’s computer. The tool is now the most widely used AI coding assistant at work, per JetBrains’ 2026 survey of 15,000 developers, and Anthropic has spent the year extending it with memory upgrades and new guardrails.
Source: Claude Code docs on cloud environments and Anthropic’s ClaudeDevs announcement, quoted in full.
0 Comments