Skip to content
Nubent

/AI-native: AWS + Claude

Claude Code for enterprise: a structured rollout for engineering teams

We help your team adopt Claude Code, Anthropic's agentic coding tool, with clear rules. We set up CLAUDE.md, permissions, MCP servers and CI/CD usage, connect the models through Amazon Bedrock if you need it, and measure the impact on delivery.

/Diagnosis

Signs it is time

Every engineer uses it differently

Some already use Claude Code and some don't, each with their own setup, and there is no agreement on what the agent is allowed to do in the repository.

Security hasn't signed off yet

There are open questions about credentials, production access, what code leaves the company and where it is processed, and nobody has time to answer them.

The agent doesn't know your codebase

Claude Code suggests changes that ignore your conventions, your architecture and your build commands, because nobody gave it that context.

You can't tell if it's paying off

You pay for seats or token usage, but you have no data on review times, pull request quality or real adoption.

/What we deliver

Concrete deliverables, in your repository

  • Team baseline configuration

    Shared settings in the repository, default permissions, allowed and blocked commands, and hooks for formatting, linting and tests.

  • A CLAUDE.md per repository

    Context files covering each repository's architecture, conventions, build and test commands, and security rules.

  • Amazon Bedrock integration

    Claude Code pointed at the Claude models in your AWS account, with IAM roles, IAM Identity Center and usage controls per team.

  • Approved MCP servers

    A catalog of reviewed MCP servers for tickets, documentation and observability, read-only by default.

  • Claude Code in CI/CD

    GitHub Actions workflows for pull request review and assisted tasks, using OIDC into AWS instead of static keys.

  • Usage guide and metrics

    Written best practices for the team, pairing sessions and a dashboard tracking adoption and delivery flow.

/How we work

Our process

  1. Discovery

    We review repositories, pipelines, security policies and how the team works today, and agree on which metrics to track.

  2. Pilot with one team

    We set up Claude Code in one or two repositories, write the CLAUDE.md and work alongside the team on real tasks.

  3. Standardization

    We roll the configuration, permissions, MCP servers and CI workflows out to the remaining repositories by pull request.

  4. Measure and adjust

    We compare metrics against the baseline, refine rules and context, and leave the team able to maintain it on its own.

What Claude Code is and why to adopt it with rules

Claude Code is an Anthropic tool that works in the terminal and the editor: it reads the repository, runs commands, edits files and proposes complete changes. An individual engineer can install it in minutes. For a company, the question is different: what the agent can do, with which credentials, on which repositories, and how its output gets reviewed.

Without those rules, everyone decides on their own. You end up with overly broad permissions, incomplete context and uneven results across teams. Our job is to turn that ad hoc adoption into an engineering practice: shared configuration, written context, clear limits and human review of everything that reaches the main branch.

CLAUDE.md, permissions and MCP servers

The CLAUDE.md file gives the agent the context a new engineer would need: architecture, conventions, build and test commands, and what it must never touch. A good CLAUDE.md is short, specific and lives in the repository, so it is reviewed and updated like any other code.

Permissions define which commands Claude Code can run without asking and which ones are blocked. We start conservative and widen them based on evidence. The same goes for MCP servers: we build an approved catalog, read-only by default, so the agent can look up tickets, documentation or observability data without opening unnecessary doors.

Claude Code with models on Amazon Bedrock

Claude Code can use the Claude models through Amazon Bedrock instead of the Anthropic API. You enable it with the CLAUDE_CODE_USE_BEDROCK environment variable and it picks up credentials from the standard AWS chain, so it fits with IAM Identity Center and the roles your team already uses.

For many companies this clears the main security objection: usage stays in the AWS account, is billed with the rest of the platform and is audited with CloudTrail. We configure model access, inference profiles, quotas and per-team cost alarms, all in Terraform.

Claude Code in CI/CD and code review

Claude Code also runs in pipelines. With GitHub Actions it can review pull requests, respond to comments or implement scoped tasks from an issue. We set up those workflows with OIDC into AWS so they use Bedrock without static keys, with least-privilege repository permissions and limits on runtime and usage.

The rule we keep is simple: the agent proposes and a person approves. No change reaches the main branch without human review, and the workflows log what the agent did. That way, assisted review speeds up the team's work without replacing the judgment of the people who know the system.

How to measure the impact of Claude Code

Counting generated lines of code says very little. We prefer flow metrics the team already understands: time from first commit to merge, time to first review, pull request size, deployment frequency and change failure rate. We take a baseline before the pilot and compare afterward.

We add adoption and usage data per team, plus short interviews with engineers to learn where it helps and where it doesn't. With that, you can decide on evidence whether to expand usage, change the rules or focus the agent on specific work such as tests, migrations or documentation.

/FAQ

Claude Code for teams: frequently asked questions

Another question? Write to hello@nubent.com and an engineer will answer.

Can Claude Code be used with Amazon Bedrock?

Yes. Claude Code supports Amazon Bedrock as a model provider. You enable it with the CLAUDE_CODE_USE_BEDROCK environment variable and it uses the standard AWS credential chain, so it works with profiles, IAM Identity Center or roles. You first need to enable access to the Claude models in the Bedrock console of your account.

Is it safe to use Claude Code on my company's code?

It can be, if it is configured properly. We set conservative permissions, block dangerous commands, keep production credentials off developer machines and use Bedrock when security requires inference to stay in your AWS account. On top of that, every change goes through a pull request and human review. We document those decisions so your security team can approve them.

What is the CLAUDE.md file?

CLAUDE.md is a text file in the repository that Claude Code reads at the start of a session. It holds the project context: architecture, conventions, build and test commands, and the rules the agent must follow. It works as an onboarding guide for the agent, and it is versioned and reviewed like the rest of the code.

Can Claude Code review pull requests on GitHub?

Yes. Anthropic publishes a GitHub Actions integration that lets Claude Code review pull requests, respond to comments and implement tasks from issues. It can use the models through Amazon Bedrock with OIDC authentication. We configure it with least-privilege permissions and always as support for human review, which still approves the merge.

How long does it take to roll out Claude Code to a team?

A pilot on one or two repositories is short, because it works on the team's real tasks. Standardizing across the rest of the repositories depends on how many there are and how mature the pipelines are. We plan it in stages, with baseline metrics, so each step has evidence behind it before the next one starts.

Next step

Let's talk about your Claude Code for teams project

30 minutes with an engineer. Concrete next steps, whether or not you work with us.