Claude Code HIPAA Setup: Prepare Every Developer Computer

To run Claude Code in a HIPAA-ready organization, you need a Claude Enterprise plan, a Business Associate Agreement (BAA) with Anthropic, and the HIPAA configuration turned on for your organization. The Primary Owner of the organization applies it. Your part, as the IT or security admin, is to get every developer's computer ready before that happens, then confirm it took effect.

The configuration covers Claude Code in local mode, which means a session running on the developer's own computer: Claude Code in the terminal, the Code tab in the Claude desktop app, and Cowork in the desktop app. Cloud sessions are not part of it. The Claude Code extensions for VS Code and JetBrains keep working with the configuration applied, but your BAA does not cover them. If your developers live in the desktop app, that is good news, because the Code tab is in scope.

This is the step most rollouts get wrong. The HIPAA configuration only takes effect in sessions where the developer signs in with a Claude Enterprise account and Claude Code talks directly to the Claude API. On any other connection, Claude Code still runs. It just runs without the HIPAA configuration.

To check one computer, start Claude Code and run /status. A correct setup shows Login method as Claude Enterprise account and your organization on the Organization line. If you see an API provider line or an Anthropic base URL line, that session is going through a cloud provider, a gateway or a custom endpoint, and it is not getting the configuration. If you set up an LLM gateway or Bedrock earlier for cost reasons, this is where the two plans collide.

The HIPAA configuration needs Claude Code v2.1.285 or later and Claude Desktop v2.19675.0 or later. Update both if your team uses both. Run claude --version in a terminal to read the Claude Code version.

Once the configuration is on, requests from older versions are rejected. In the terminal, every request fails with an API error saying the version is older than your organization's policy allows. In the desktop app, the developer gets an Update required dialog before they can keep using the Code tab. The minimum version rises over time, so keeping auto-updates working matters more here than anywhere else.

Allow these hosts through your proxy and firewall over HTTPS on port 443. Allow the whole host, not individual paths. A corporate HTTPS proxy is fine, as long as it can reach them. If your proxy inspects TLS, the corporate proxy guide covers trusting its certificate.

The api.anthropic.com line is the one people miss. Claude Code learns your organization has the HIPAA configuration by fetching a policy from that host when it starts, and again about every hour while the session is in use. Block it and the restrictions never arrive. The desktop app and npm or Homebrew installs need a few more hosts, so check those lists too if your team installs that way.

Managed settings are how you stop developers from drifting onto a connection the configuration does not cover. A good starting point sets four keys in the managed settings file you already deploy. They work whether or not the HIPAA configuration is applied yet, so you can roll them out early.

Then confirm the file loaded. Run /status on a managed computer: the Setting sources line should list Enterprise managed settings, and the Allowed providers line should read Anthropic API (managed allowedProviders). For a fuller file with sandboxing, a network allowlist and credential protections, the Claude Code repository ships a settings-hipaa.json example next to its other settings examples.

My recommendation: treat requiredMinimumVersion as part of the HIPAA rollout, not an optional extra. It is the only thing that stops an old install from quietly starting on a cloud provider after you thought you had locked it down.

If HIPAA is missing from /status, check three things in order. First the account and connection: no API provider or base URL line should appear. Second the policy fetch: the Organization policy line in /status, or claude doctor outside a session, says where the policy came from or why it failed, which usually means api.anthropic.com is blocked. Third, ask the Primary Owner whether they have actually applied it.

Two of these deserve a heads up to your team. MCP tool search being off means every connected MCP tool loads into context at the start, so trim the servers people do not need. And the credential stripping only covers Anthropic credentials. Cloud provider and GitHub credentials are left alone, so a command that pushes to GitHub still works, and the BAA does not cover what it sends there. Use permission rules and the sandbox to limit which commands and hosts Claude can reach.

Local mode means session data lives on each developer's computer, and securing and deleting it is your organization's job. Transcripts for both the terminal and the Code tab sit in ~/.claude/projects/. The desktop app also keeps a data folder, at ~/Library/Application Support/Claude on a Mac. With the configuration on, the desktop app deletes Code tab sessions inactive for longer than cleanupPeriodDays, starred ones included, but only while the app is running.

The same catch applies in the terminal: the retention sweep only runs when someone starts Claude Code. A computer nobody opens keeps its data. To delete a developer's session data right away, sign in as that developer and run claude purge --all --yes on v2.1.288 or later, or claude project purge --all --yes on v2.1.126 through v2.1.287. That removes transcripts, auto memory, history and project entries, but it leaves a few paths such as pasted text in paste-cache, so it is not a full wipe.

Offboarding is where this bites. Removing a seat deletes nothing on the computer, and /logout does not delete session data either. If a developer leaves, wipe the machine with your device management tool before it is reassigned.