Claude Code for enterprise: features and admin controls
Claude Code for enterprise teams: choosing an API provider, managed settings, what admins can lock down, usage analytics, and data handling.
A platform team gets the go-ahead to give 300 engineers Claude Code. The questions arrive the same day: Can we stop people adding random MCP servers? Who's using it? Does our code train the model? Can we pin a version? Every one has an answer, spread across a dozen pages of docs.
This guide pulls them together from Anthropic's admin setup guide, checked on 16 November 2026. It doesn't cover pricing, which changes; check with Anthropic for that.
The five decisions
Anthropic's guide orders deployment into five decisions:
| Decision | What you're choosing |
|---|---|
| API provider | Where Claude Code authenticates and how it's billed |
| Settings delivery | How managed policy reaches developer machines |
| Enforcement | Which tools, commands, and integrations are allowed |
| Usage visibility | How you track adoption and spend |
| Data handling | Retention and compliance posture |
SSO, SCIM provisioning, and seat assignment are configured at the Claude account level, not in Claude Code.
1. Choose an API provider
| Provider | Choose when |
|---|---|
| Claude for Teams or Enterprise | You want Claude Code and claude.ai under one per-seat subscription, no infrastructure. Anthropic's default recommendation |
| Claude Console | API-first, pay-as-you-go |
| Amazon Bedrock | Inherit AWS compliance controls and billing |
| Google Cloud's Agent Platform | Inherit GCP controls and billing |
| Microsoft Foundry | Inherit Azure controls and billing |
Some features need a claude.ai account: cloud sessions, cloud scheduled tasks, Code Review, and Remote Control among them. Check the feature list before choosing a cloud provider.
2. Deliver managed settings
Managed settings are organization policy that takes precedence over developers' own configuration, with a few security-sensitive exceptions. Claude Code checks four sources, highest priority first:
| Mechanism | Delivery |
|---|---|
| Server-managed | The claude.ai admin console, fetched at startup and refreshed hourly |
| plist or registry policy | macOS com.anthropic.claudecode plist, Windows HKLM registry |
| File-based | managed-settings.json in a system path per OS |
| Windows user registry | HKCU; a convenience default, not enforcement, since users can write it |
The admin console route needs a Teams or Enterprise plan. Mixed-provider organizations can combine server-managed settings with a file or registry fallback. See Claude Code managed settings for examples.
3. Decide what to enforce
What managed settings can control, from the guide:
- Permissions: allow, ask, or deny tools and commands; make managed rules the only source; disable
--dangerously-skip-permissions - Starting permission mode: choose it, or remove auto mode
- Sandboxing: OS-level filesystem and network isolation with a domain allowlist
- Organization CLAUDE.md: guidance deployed to every machine
- MCP servers: allowlists, denylists, or a fixed managed set
- Plugins: restrict marketplace sources and block sideloading
- Customization lockdown: block skills, agents, hooks, and MCP servers from user and project sources
- Hooks: allow only managed hooks, restrict HTTP hook URLs
- Login and providers: restrict login method, organization, or allowed API providers
- Models and effort: limit available models, cap effort levels
- Versions: a minimum version floor, or a required version range
- Telemetry: turn off non-essential Anthropic-bound traffic
The guide flags one gap to close: denying the web fetch tool doesn't stop curl if Bash is allowed. The sandbox's network allowlist does.
4. Set up usage visibility
| Capability | Availability |
|---|---|
| OpenTelemetry export of sessions, tools, and tokens | All providers |
| Analytics dashboard with adoption and contribution metrics | Teams and Enterprise at claude.ai; Console has per-user usage and spend |
| Analytics API | Enterprise Analytics API; Claude Code Analytics API for Console |
| Spend and rate limits | Organization settings, workspace limits, or a gateway |
For request-level audit logs, Anthropic suggests a gateway between developers and the provider; its self-hosted Claude apps gateway records a per-request audit log with identity from your IdP.
5. Review data handling
Anthropic's docs say that on Team, Enterprise, Claude API, and cloud provider plans, it doesn't train models on your code or prompts. Your provider sets retention. Qualified Enterprise accounts can get Zero Data Retention, and Enterprise organizations with HIPAA enabled get a configuration with some features off. Note that cloud sessions and Code Review aren't available under Zero Data Retention or the HIPAA configuration.
Verify and roll out
Have a developer run /status: the Setting sources line should show enterprise managed settings and which source won. Then roll out in stages. See how to roll out Claude Code to an engineering team.
What policy can't distribute
Managed settings make every session follow the same rules. They don't make every session know the same decisions: which service is being deprecated, which vendor was chosen last month, which API is frozen. A managed CLAUDE.md can hold stable guidance, but decisions change weekly and need an owner.
Where Lockstep fits
Lockstep distributes team decisions the way managed settings distribute policy. It captures decisions from PRDs, Slack, and pull requests, routes each to its owner for approval, and briefs every Claude Code session on the ones that apply. Its Standards & Skills feature rolls Claude Code skills out by team, repository, path, or task type. Briefings and managed skills work in Claude Code only.
See how it works.
Sources checked 16 November 2026
- https://code.claude.com/docs/en/admin-setup
- https://code.claude.com/docs/en/managed-settings
- https://code.claude.com/docs/en/claude-code-on-the-web
Related reading
For permission rules, see Claude Code permissions. For choosing between front ends, see Claude Code desktop vs CLI vs VS Code. For vulnerability scanning, see Claude Code security.
Frequently asked questions
- What does Claude Code offer for enterprises?
- Central policy through managed settings that override local configuration, controls over permissions, MCP servers, plugins, hooks, models, and versions, usage analytics and OpenTelemetry export, and data handling options such as no training on your code on Team, Enterprise, API, and cloud provider plans, plus Zero Data Retention for qualified Enterprise accounts.
- Which API provider should an enterprise use for Claude Code?
- Anthropic's admin guide recommends Claude for Teams or Enterprise by default: Claude Code and claude.ai under one per-seat subscription with no infrastructure. Choose Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry to inherit that cloud's compliance controls and billing. Some features, such as cloud sessions and Code Review, need a claude.ai account.
- How do admins enforce Claude Code settings?
- With managed settings, delivered from the claude.ai admin console (server-managed, highest priority), macOS plist or Windows registry policy, or a managed-settings.json file. Managed values take precedence over user and project settings, with a few security-sensitive exceptions.
- Can admins see how Claude Code is used?
- Yes. Team and Enterprise plans have an analytics dashboard with adoption and contribution metrics, Enterprise has an Analytics API, and every provider supports OpenTelemetry export of sessions, tools, and tokens. Per-user spend on Team and Enterprise comes from the spend report in organization analytics settings.
- Does Anthropic train on enterprise code from Claude Code?
- Anthropic's docs say that on Team, Enterprise, Claude API, and cloud provider plans, it doesn't train models on your code or prompts. Retention and compliance depend on your API provider, and Zero Data Retention is available to qualified Enterprise accounts.
Keep reading