Claude's Projects beta now tracks dependencies and API keys in cloud coding sessions

What Anthropic is testing

Anthropic is trialing an upgraded version of Projects, the part of Claude that organizes ongoing work. As Crypto Briefing reported, the beta lets Claude work out which dependencies a project needs and which API keys it must have while running in Claude Code cloud sessions. Claude Code is Anthropic's coding assistant, and a cloud session means the work runs on Anthropic's servers rather than on your own laptop.

A dependency is an outside piece of software, such as a code library or online service, that your project needs in order to run. An API key is a secret password-like string that lets your software use another company's service. If either is missing, a project can stall partway through.

How the upgraded Projects works

The pitch is less hand-holding on big jobs. Claude can define the scope of a project, divide the work into parallel threads running in the cloud, and keep track of how those pieces rely on each other. When the threads finish, Claude looks over what they produced.

It also handles the API keys and environment settings a project needs during Claude Code sessions. Environment settings are the configuration values a program reads when it starts up.

The practical gain is spotting requirements early. If Claude flags a missing library or credential at the start, you avoid discovering it an hour into a long session.

Who gets it, and what it costs

The beta is for subscribers on the Pro, Max and Ultra plans, and Claude Code access is now bundled under those three tiers. The report puts the start of the rollout around mid-to-late September 2026, as part of a wider overhaul Anthropic began that month.

• Pro: $20 per month, or $17 per month when billed annually.
• Max: $100, with five times the usage of Pro.
• Ultra: $200 per month, with ten times Pro's weekly usage.

Ultra users who have been verified no longer hit the five-hour session limit. The report says that change was confirmed as of October 30-31, 2026. Nothing is said about any charge beyond the plan itself.

What changes for existing users

People already paying for Claude Code don't have to rebuild anything. Anthropic is moving existing projects to the new system by itself, and the report says that for some paying users this happens without losing any stored data. The wording is limited to "some users," so it is not a blanket guarantee.

The Ultra change speaks to heavy users. Agent tasks that run for hours don't fit well inside short capped windows, and dropping the cap for verified accounts suggests Anthropic knows it.

The larger direction is autonomy. Instead of doing tasks one after another, Claude can hand pieces to several threads and gather the results. For developers, that means directing multiple streams of work without checking every step.

The question the report leaves open

The coverage doesn't say how Claude stores the API keys it tracks, who can see them, or how they are protected. That gap is the main thing to settle. A feature that finds dependencies is a time-saver, but one that holds credentials is a security matter.

This is also coverage from a secondary outlet, and such write-ups sometimes smooth over product details. Anthropic's own release notes are the better place to confirm what is actually switched on for your plan.

Until those details are clear, the sensible approach is caution: use keys that are limited in scope or disposable, and don't hand over credentials that guard anything you can't afford to rotate.