Claude Code enterprise setup

Claude Code hit a ~$2.5 billion annualized run rate with 5.2 million VS Code installs by early 2026, ranking first in the Pragmatic Engineer's developer survey. Your engineers are already using it. The question is whether they're all using the same configuration, the same context, and the same guardrails, or whether each developer has built their own local setup that diverges a little more every week. This page covers what breaks in an unmanaged rollout and what a properly deployed team setup looks like.

What breaks when Claude Code has no shared configuration#

Configuration drift: every engineer running a different setup#

When engineers adopt Claude Code individually, they each make reasonable local decisions: this CLAUDE.md covers the project, this MCP server connects to the database, this permission set feels right. Multiply that by twenty engineers and you have twenty different configurations, with different context, different tool access, and different expectations from the same AI assistant. Productivity gains are real but uneven. Some engineers are getting substantial help; others are running a generic coding assistant because their setup never got past the defaults.

Configuration drift is invisible until it causes a problem. Then it causes it repeatedly, across different engineers' machines, for different reasons.

No codebase context: Claude as a generic assistant instead of a team tool#

Without MCP servers connecting Claude Code to your internal repos, API documentation, database schemas, and architecture decisions, engineers are feeding Claude context manually on every session: pasting code, explaining architecture, re-establishing what it already knows. That overhead eats a large part of the productivity benefit.

The difference between a Claude Code setup with proper codebase context and one without is roughly the difference between a senior engineer who knows your stack and a capable hire who just started. The tool gives you the latter by default. Getting to the former requires infrastructure.

Compliance gaps: no audit logs, no tool permissions, no governance#

Claude Code Enterprise adds SSO, SCIM, audit logging, managed policy settings, and a Compliance API. The individual-account tier includes none of these. Without a managed deployment, engineering teams using Claude Code on individual accounts have no way to define which tools the agent can run, no audit trail of actions taken during agentic workflows, and no centralized policy enforcement. For teams under SOC 2, ISO 27001, or any framework that requires AI action traceability, that's a gap that needs to close before the tooling is production-sanctioned.

What we set up#

Claude Code enterprise account and admin controls#

We configure your Claude Code Enterprise account, provision admin roles, and set up the managed policy infrastructure: SSO integration (SAML/OIDC), SCIM for user provisioning, and the admin dashboard for tool permission management. Enterprise accounts include the Compliance API for audit log export to your SIEM.

CLAUDE.md authoring: team standards encoded in the repo#

The CLAUDE.md file is how Claude Code learns your team's conventions: naming patterns, architecture decisions, preferred libraries, review standards, security constraints. We author CLAUDE.md files at both the repository and project level, encoding your actual engineering standards rather than generic prompts. These files live in your repo, version-controlled alongside your code, and become the shared context baseline every engineer on your team inherits automatically.

Getting this right is mostly an interview process. We go through your codebase and talk to your engineering lead about what decisions are already made, what's contested, and what tripped up the last few engineers who joined. That's what ends up in the file.

Custom MCP servers for your internal codebase and APIs#

MCP server development is the part of the engagement that takes the most time and makes the biggest difference. Claude Code's enterprise governance system lets admins deploy a managed-mcp.json to system-wide directories, which centrally controls which MCP servers every Claude Code instance connects to. We build the MCP servers that make that infrastructure worthwhile: connections to your internal repos, your API documentation systems, your database schemas, and any internal knowledge bases or architecture decision records your engineers reference regularly. Every engineer gets the same codebase context without doing any setup themselves.

Custom skills for your team's recurring workflows#

Custom skills in Claude Code use a SKILL.md file with a YAML frontmatter name field that becomes a slash-command. On Enterprise plans, organization Owners can provision skills org-wide so they appear automatically for all users. We build skills for your team's recurring patterns: code review workflows, test generation, migration scripts, PR description generation, architectural analysis. Repeated tasks become single-command operations rather than prompt-from-scratch exercises.

CI/CD pipeline integration#

We integrate Claude Code's agentic capabilities into your CI/CD pipeline where it makes sense: automated code review on pull requests, test generation hooks, documentation generation on merge. What this actually looks like varies by team. Some teams want automation light and use CI integration mostly for documentation. Others want heavier review automation. We scope it during the assessment rather than prescribing a pattern upfront.

Team training on agentic coding patterns#

Configuration alone doesn't change how engineers work. We run hands-on training sessions covering how to write effective prompts for agentic tasks, how to use CLAUDE.md to set context efficiently, how to invoke custom skills, and how to work with Claude Code on multi-step tasks without getting into approval loops. Training is tailored to your team's actual codebase and workflows, not generic demos.

Custom MCP servers: where consistent productivity comes from#

What MCP servers do and why shared config depends on them#

MCP (Model Context Protocol) servers are the interfaces that give Claude Code access to specific data sources and tools. An MCP server for your internal repos means Claude Code can read your actual codebase rather than working from manually pasted excerpts. An MCP server for your API documentation means it can answer questions about your own services without engineers copying and pasting specs into the chat.

Without this, Claude Code is a good generic coding assistant. That's not useless, but it's also not what the productivity numbers are based on. The gap between "good generic assistant" and "knows your codebase" is where most of the value lives, and it takes infrastructure to close.

Typical builds: repo context, API docs, internal knowledge bases#

Common MCP servers we build for enterprise Claude Code deployments:

• Repository context server: indexes your codebase for Claude Code to query by file, module, or function
• API documentation server: serves your internal API specs, including endpoints, schemas, and usage examples
• Architecture decisions server: surfaces ADRs and design documents relevant to the current task
• Database schema server: gives Claude Code read access to your schema definitions for query and migration work
• Issue and ticket server: connects Claude Code to your project management system for context on the current task or ticket

The actual priority depends on your team. Codebase context is almost always first. After that it varies.

Managed MCP governance for enterprise environments#

With the managed-mcp.json deployment feature, enterprise admins control which MCP servers every engineer's Claude Code instance connects to, with no per-user setup required. Engineers get the right context by default. New MCP servers roll out centrally. Deprecated connections are removed without engineers having to touch their own configs. This is what makes consistent context sustainable rather than something you set up once and watch diverge.

Claude Code enterprise vs. individual accounts#

What the enterprise tier unlocks#

The Claude Code Enterprise tier adds: SSO (SAML/OIDC), SCIM user provisioning, organization-wide skill provisioning, managed MCP deployment via managed-mcp.json, admin policy controls, usage analytics, and the Compliance API for audit log export. The individual account tier includes none of these.

Anthropic's Enterprise plan pricing is custom, contact required. The Team plan (which includes Claude Code) is $150/seat/month on the Premium tier (Anthropic, 2026).

How team-level configuration differs from individual setup#

Individual setup is local. Team-level configuration is infrastructure. When an engineer leaves and their replacement joins, team-level configuration means the new hire inherits the full context immediately. CLAUDE.md files are in the repo. MCP servers are centrally managed. Custom skills are org-provisioned. Nothing depends on the previous engineer's local setup surviving.

That last point matters more than it sounds. In practice, a departing engineer's local Claude Code setup is gone the day they leave. Team-level configuration is what makes institutional knowledge persist.

Security, SSO, and compliance considerations#

Enterprise Claude Code integrates with your identity provider via SAML/OIDC, so onboarding and offboarding run through your existing IAM workflow rather than requiring separate Claude Code account management. SCIM automates user provisioning from your directory. The Compliance API exports audit logs in a format your SIEM can ingest. For teams under SOC 2, these are the controls that make Claude Code a sanctioned enterprise tool rather than a productivity experiment.

How the engagement works#

Step 1: codebase and workflow assessment#

We spend time with your engineering lead to understand your codebase structure, your team's current Claude Code usage patterns, the tools and systems that would benefit most from MCP connections, and your compliance requirements. This produces the engagement spec: which MCP servers to build, what CLAUDE.md standards to encode, which skills to create, and what CI/CD integrations to target. Most teams have a clearer picture of what they want by the end of this conversation than they did going in.

Step 2: MCP server build and CLAUDE.md configuration#

MCP server development is the core of the engagement. Each server is built to your spec, tested against your actual systems, and validated with a small group of engineers before being promoted to the managed config. Things come up during this phase that weren't obvious in assessment, usually around authentication or access scope. We work through those with your team rather than treating them as blockers. CLAUDE.md files are drafted, reviewed with your engineering lead, and committed to your repos.

Step 3: custom skills and CI/CD integration#

Skills are built as SKILL.md files and provisioned org-wide through your Enterprise admin panel. CI/CD integrations are implemented in your pipeline and validated on a staging branch before being merged to main.

Step 4: team training and handoff#

We run training sessions with your full engineering team: typically two 90-minute sessions covering agentic workflows, skill usage, and CLAUDE.md management. Handoff documentation covers every component built, how to extend the configuration, and how to manage MCP servers and skills going forward. The goal is that your team can maintain and extend this without coming back to us.

Engagements typically complete in 2 to 4 weeks depending on MCP server count and CI/CD integration complexity.

FAQ#

How much does a Claude Code enterprise setup cost? Engagements start at $5,000 per team and scale with the number of MCP servers built, CI/CD integrations, and training sessions required. Most teams in the 10 to 50 engineer range fall in the $5,000 to $15,000 range. Scope is confirmed after the assessment.

What is a CLAUDE.md file and how do teams use it? A CLAUDE.md file is a Markdown document placed in your repository that Claude Code reads automatically when starting a session in that directory. It encodes your team's conventions: architecture patterns, naming standards, preferred libraries, security constraints. Every engineer gets this context without having to specify it themselves. The file is just Markdown checked into your repo, so it versions alongside your code and can be reviewed and updated like any other file.

How do you build a custom MCP server for Claude Code? MCP servers are small server processes that expose a defined set of tools and resources via the Model Context Protocol. We build them to connect Claude Code to your specific internal systems, typically using your existing APIs and authentication mechanisms. Each server is documented and maintained as part of your engineering infrastructure. Build time per server is usually 3 to 5 days, depending on how straightforward your APIs are to work with.

What does Claude Code enterprise include that individual accounts don't? SSO, SCIM, org-wide skill provisioning, managed MCP deployment via managed-mcp.json, admin policy controls, usage analytics, and the Compliance API for audit log export. The short version: everything that makes Claude Code governable at a team level rather than just useful to individuals.

How long does it take to deploy Claude Code for an engineering team? 2 to 4 weeks is typical. MCP server development is the longest phase, usually 1 to 2 weeks depending on integration count. CLAUDE.md authoring, skill builds, and training can run in parallel with MCP development. Teams with complex authentication setups or large numbers of internal systems can run longer.

Request an audit and we'll review your current Claude Code setup, or lack of one, and give you a concrete picture of what a team deployment would cover. If you haven't yet settled on which tools belong in your stack, AI coding stack selection covers that. The AI developer tooling hub has the full services overview.