Claude Code on Amazon Bedrock: when a team should route through AWS

Last updated: November 12, 2026

TL;DR

Route Claude Code through Amazon Bedrock when AWS has to be the party that processes your prompts and code, or when your compliance program needs FedRAMP High, IL4, IL5 or HIPAA readiness on AWS. If the real goal is only to pay through AWS, Claude Platform on AWS keeps more of Claude Code's features, and for most teams Anthropic's own docs recommend a Claude Team or Enterprise plan instead.

• 🏛️ Amazon Bedrock: AWS runs inference and is the data processor, with IAM, CloudTrail, Guardrails and Cost Explorer. You give up every feature that needs a Claude subscription, plus web search, fast mode, the Advisor, Channels, the analytics dashboard and server-managed settings.
• 🧾 Claude Platform on AWS: Anthropic runs it and AWS Marketplace bills it. New API features typically land the same day as on the Claude API, web search works, and GitHub Actions doesn't.
• 👥 Claude Team or Enterprise: the default Anthropic recommends for most organizations. It adds Claude on the web, SSO, the analytics dashboard, server-managed settings, Remote Control and cloud sessions.
• 📌 On any AWS route: pin model versions and lock the provider in managed settings before you roll out.

Three ways to run Claude Code with AWS in the picture

The deciding question is who operates the model and who processes your data, not where the invoice lands. Anthropic's Claude Platform on AWS docs and Claude Code's enterprise deployment overview set the three options side by side:

Anthropic's docs are direct about the Bedrock case: organizations "that require FedRAMP High, IL4, IL5, or HIPAA readiness, or that need AWS to be the sole data processor" should use Bedrock, which "runs entirely on AWS-controlled infrastructure with AWS as the operating party" (Claude Platform on AWS docs). If neither of those is your reason, keep reading before you commit.

What a team gives up on Bedrock

Everything that runs locally works on every provider: the CLI, the VS Code and JetBrains extensions, subagents, hooks, skills, CLAUDE.md, plugins, MCP servers, checkpoints, sandboxing, OpenTelemetry metrics and the managed settings file (feature availability). What you lose sits on Anthropic's side of the wire.

Not available on Bedrock, per the same page:

Auto mode works on Bedrock only with Sonnet 5 or later, Opus 4.7 or later, Haiku 5.5 and the Fable models. Cross-session messaging works only between sessions on the same machine.

Two of those gaps change how an admin works. Without server-managed settings, policy has to reach each machine through the managed settings file, which our Claude Code settings guide covers. Without the analytics dashboard, usage visibility comes from OpenTelemetry and your AWS bill, the setup our Claude Code monitoring guide walks through.

What a team gets on Bedrock

The gains are AWS's own controls, applied to Claude Code:

• AWS as the data processor. On Bedrock, AWS is the inference data processor, where on Claude Platform on AWS and Anthropic's own plans it is Anthropic.
• IAM and CloudTrail. Access runs through IAM policies, and calls show up in CloudTrail. Claude Code's Bedrock guide lists the four bedrock: actions a role needs and suggests restricting the resource to specific inference profile ARNs.
• Guardrails. Amazon Bedrock Guardrails can filter Claude Code's traffic once you add the guardrail headers to your settings file.
• Service tiers. ANTHROPIC_BEDROCK_SERVICE_TIER picks default, flex or priority, trading cost against latency, and provisioned throughput covers reserved capacity.
• Quieter defaults. Error reporting, telemetry and bug reporting to Anthropic are off by default on Bedrock. Session quality surveys and the WebFetch domain safety check still run unless you turn them off (data usage).
• The full context window. Fable models, Sonnet 5 and later, and Opus 4.7 and later run with the 1M token context window by default on Bedrock, with no model suffix needed.

The Bedrock guide also suggests a dedicated AWS account for Claude Code, so cost tracking and access control don't mix with the rest of your AWS spend.

Claude Platform on AWS: the middle option

Claude Platform on AWS is the Anthropic-operated Claude API with AWS authentication, IAM access control and AWS Marketplace billing (Claude Code docs). Requests reach Anthropic's API directly, so you get the same models and API features on the same release schedule.

It keeps two things Bedrock doesn't: web search, and the Advisor on Claude Code v2.1.300 or later. It drops one thing Bedrock keeps: GitHub Actions, though GitLab CI/CD works. Fast mode, Channels, the analytics dashboard and server-managed settings are missing here too.

Three details matter before you sign up:

• It's a new organization. Subscribing through AWS Marketplace provisions a new Anthropic organization tied to your AWS account. API keys, workspaces and Console settings from an existing Anthropic organization don't carry over.
• Anthropic processes the data. Anthropic is the data processor for inference, running on AWS infrastructure. For workspaces created before September 18, 2026, Anthropic might process requests outside AWS, and the docs point to the inference_geo parameter to pin a geography.
• It's a separate capacity pool from both the Claude API and Bedrock, so you can run on more than one and fail over between them.

Which one should your team choose?

If you are weighing a team plan against Enterprise at all, our Claude Team vs Enterprise comparison covers what the bigger plan adds. If the worry behind the Bedrock question is what Claude Code can reach, start with is Claude Code safe for a business, because the provider changes who processes data, not what Claude Code is allowed to touch on a developer's machine.

Setting up Claude Code on Bedrock for a team

One developer can run claude, pick 3rd-party platform, then Amazon Bedrock, and let the wizard write the settings. Anthropic's guide says a team rollout should use the manual setup and pin models first. The steps that matter:

1. Submit the use case form once. Each AWS account submits use case details in the Bedrock console before its first Anthropic model call. With AWS Organizations, submit it once from the management account through the PutUseCaseForModelAccess API, and approval extends to child accounts.
2. Write a narrow IAM policy. Grant bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream, bedrock:ListInferenceProfiles and bedrock:GetInferenceProfile, scoped to the inference profiles you approve.
3. Pin model versions. Set ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL, ANTHROPIC_DEFAULT_HAIKU_MODEL and ANTHROPIC_DEFAULT_FABLE_MODEL. Without pins, aliases resolve to Claude Code's built-in default for the provider, which "can lag the newest release and may not yet be enabled in your account" (enterprise deployment overview).
4. Lock the provider. Set allowedProviders to ["bedrock"] in managed settings so a developer can't switch a managed machine to another provider. It needs Claude Code v2.1.285 or later.
5. Add the optional controls. A Guardrail, a service tier, and a dedicated AWS account for Claude Code, if your security and finance teams want them.

Our Claude Code permissions guide covers the rules that decide what Claude Code may do once it's running, which stay the same on every provider.

When Bedrock is the wrong call

Bedrock solves a data-processing and procurement problem. It's the wrong call if your developers rely on Remote Control, cloud sessions, Code Review or the analytics dashboard, or if they also want Claude on the web, because none of those come with it. It's also more setup than the alternatives for a small team: IAM, model pins and an AWS account to watch, where a team plan is a sign-in. For the broader rollout questions, our Claude Code best practices for a team is the place to start.

Frequently asked questions

Can Claude Code use Amazon Bedrock?

Yes. Set CLAUDE_CODE_USE_BEDROCK to 1 and give Claude Code AWS credentials, or run claude and choose 3rd-party platform, then Amazon Bedrock, to use the setup wizard. Your AWS account first submits a one-time use case form in the Bedrock console.

What Claude Code features don't work on Bedrock?

Every feature that needs a Claude subscription, such as cloud sessions, Remote Control, Code Review and Slack, plus web search, fast mode, the Advisor, Channels, the analytics dashboard, server-managed settings and the design-sync and import commands. The CLI, IDE extensions, hooks, skills, subagents and MCP servers all work.

Does Claude Code on Bedrock send data to Anthropic?

On Bedrock, AWS operates inference and is the data processor. Claude Code also turns off error reporting, telemetry and bug reporting to Anthropic by default there. Session quality surveys and the WebFetch domain safety check still run unless you disable them.

What's the difference between Claude Platform on AWS and Amazon Bedrock?

Bedrock is operated by AWS, which processes your data. Claude Platform on AWS is the Claude API operated by Anthropic, with AWS sign-in, IAM and Marketplace billing. It gets new features about the same day as the Claude API and supports web search, but not GitHub Actions.

Does web search work in Claude Code on Bedrock?

No. On Bedrock, Claude Code can fetch a specific URL with the WebFetch tool but can't run a web search. Web search works on Claude Platform on AWS, the Anthropic Console, Claude plans and Microsoft Foundry, and on Google Cloud for recent models.

Can I stop developers from switching away from Bedrock?

Yes. Set allowedProviders to bedrock in managed settings, which needs Claude Code v2.1.285 or later. A managed machine then can't use another provider. Deliver it through your device management, since server-managed settings aren't available on Bedrock.

Get Claude Code set up properly for your team

We set up Claude Team for small businesses: the seat plan, the security settings, shared Projects and skills written from your actual processes, and managed Claude Code settings for teams with developers. If you'd rather have it done once and done right, see how our Claude Team setup works or book a 30-minute call.