Claude Enterprise: Features, Limits, and What It Leaves for You to Build
What Claude Enterprise includes (SSO, SCIM, audit, expanded context), how it differs from Team, and the organizational memory layer it leaves open.
TL;DR
Claude Enterprise is Anthropic's top organizational plan: the Claude apps plus Claude Code for a whole company, with SSO and SCIM provisioning, domain capture, role-based permissions, audit logs, an expanded context window and admin controls over connectors and data retention. It answers the questions security review asks. What it deliberately is not: a shared organizational memory. Each user's Projects and memory stay scoped to that user, so a hundred-seat deployment is a hundred private assistants, not one company that remembers. This guide covers what the plan includes, how it differs from Team, and the layer it leaves open.
What does Claude Enterprise actually include?
The capabilities Anthropic lists for the Enterprise plan, grouped by who asked for them:
1. For security review: SSO and SCIM. Single sign-on through your identity provider, automated provisioning and deprovisioning, and domain capture so accounts on your domain fall under the org.
2. For admins: role-based permissions and audit logs. Fine-grained roles, compliance-oriented logging of usage, and centralized control over which integrations and connectors members may enable.
3. For legal: data controls. Enterprise data is excluded from model training by default, with configurable retention. This is the clause that unblocks most procurement conversations.
4. For users: more capacity. Enterprise carries an expanded context window (Anthropic advertises 500K tokens on Enterprise seats), higher usage limits than Team, and access to Claude Code under the same governance umbrella.
5. For rollout: Projects and shared chat surfaces. Teams can share Projects and their knowledge bases within the workspace, which is genuine collaboration, scoped to what someone deliberately uploaded.
Anthropic sells Enterprise by sales contact rather than a published per-seat price, with Team (publicly priced per seat) as the self-serve tier below it.
How is Enterprise different from Team?
| Dimension | Claude Team | Claude Enterprise |
|---|---|---|
| Purchase path | Self-serve, published per-seat price | Sales contact |
| SSO / SCIM | No | Yes |
| Domain capture | No | Yes |
| Audit logs | No | Yes |
| Context window | Standard | Expanded (advertised 500K) |
| Connector governance | Per-user | Admin-controlled |
| Training on your data | Off by default | Off by default, contractual |
The pattern is the standard enterprise-software split: Team gives a group the product, Enterprise gives the org control over the product. If your security team has a vendor questionnaire, you are buying Enterprise; the only real question is seat count.
What does Claude Enterprise not do?
The boundary that matters for anyone planning an AI rollout, stated without spin:
1. Memory is per-user, not per-company. Claude's memory features and Projects belong to individual users and specific workspaces. When an engineer's Claude learns your deployment process, the PM's Claude has not learned it. Nothing aggregates what the organization collectively told its assistants.
2. Uploaded knowledge is static. A Project knowledge base holds what someone uploaded, as of when they uploaded it. It does not know the policy changed last week, and two Projects can hold contradicting versions of the same document with no system noticing.
3. Context does not cross tools. What Claude learns in conversation does not reach your coding agents, support agents or any non-Claude system, and what those systems know does not reach Claude, except through integrations you wire yourself, typically over MCP.
4. No organizational source of truth. Enterprise governs access to Claude. It does not attempt to maintain what is currently true across your company, who decided it, or when it changed. That is a different product category.
None of this is a criticism; Anthropic drew the product line deliberately, and MCP exists precisely so external systems can fill it. The practical consequence: companies deploying Claude Enterprise at scale end up pairing it with an organizational memory layer. Sentra serves that layer, one governed bi-temporal store of company facts that every seat's Claude, and every non-Claude agent, reads over MCP, so the hundredth employee's assistant knows what the first ninety-nine already established.
What should you negotiate and check before buying?
1. Seat minimums. Enterprise carries a seat floor, so small teams should price Team plus a self-managed rollout against it.
2. Usage headroom. Claude Code under heavy agentic use consumes limits faster than chat; ask how limits behave at your expected coding-agent volume.
3. Connector policy. Decide which MCP servers and integrations you will allow before rollout, because retrofitting governance after employees have wired personal connectors is painful.
4. The memory question. Ask in the eval, not after: when two departments need the same answer, where does it live? If the answer is "each uploads a copy," you have discovered the layer you still need to build or buy.
FAQ
How much does Claude Enterprise cost?
Does Claude Enterprise include Claude Code?
Is my company's data used for training on Enterprise?
Does Claude Enterprise give the whole company a shared memory?
Claude Enterprise vs ChatGPT Enterprise, which is better?
The decision rule
Buy Claude Enterprise when the blockers are governance: SSO, audit, data controls, admin-managed rollout. Just do not confuse governed access with shared knowledge. If the goal is a company where the second person to ask a question gets the first person's answer, that layer is not on this rate card, and planning for it at rollout costs a fraction of retrofitting it after every team has built its own private context.