Claude Platform on AWS: Not Bedrock — What Developers Need to Know

Anthropic launched Claude Platform on AWS in May 2026, a managed API operated by Anthropic rather than AWS, offering IAM authentication, CloudTrail logging, and advanced features like Claude Managed Agents and MCP Connector that Amazon Bedrock lacks. Developers now have three distinct paths to run Claude on AWS—Bedrock, Claude Platform, or direct API—with significant implications for compliance, data processing, and feature access.

Anthropic shipped its own managed Claude API directly through AWS in May 2026, and most developers either missed it or assumed it was just another Bedrock alias. It is not. Claude Platform on AWS is a fundamentally different product from Amazon Bedrock — operated by Anthropic, not AWS — and the gap in features and architecture is significant enough that new Claude integrations on AWS now have a real choice to make. Three Ways to Run Claude on AWS And Why It Matters Which You Pick As of mid-2026, there are three distinct paths to Claude on AWS: Amazon Bedrock — AWS operates inference; AWS is the data processor; data stays inside the AWS boundary Claude Platform on AWS — Anthropic operates inference; Anthropic and AWS are independent data processors; billed through AWS Marketplace Direct Anthropic API — entirely outside the AWS ecosystem; separate invoice The distinction between Bedrock and Claude Platform is not cosmetic. On Bedrock, AWS is processing your data. On Claude Platform on AWS, Anthropic is. Your contracts, compliance posture, and vendor agreements are affected differently. This is the primary reason the choice exists at all — compliance teams at regulated companies need to know exactly who touches their data. IAM Authentication: The Actual Developer Win The loudest benefit of Claude Platform on AWS is not the feature set — it is the authentication story. Claude Platform uses AWS IAM with SigV4 request signing, the same mechanism you use for S3, Lambda, and every other AWS service. Your existing credential provider chain just works. - No separate Anthropic API key to store in Secrets Manager, rotate, or audit independently - Claude inference calls appear in CloudTrail automatically — no extra instrumentation - IAM policies can restrict which roles, services, or environments can call Claude - Your security team already knows how to audit IAM For teams that already run governance and access control through IAM, this removes an entire layer of API key management. The endpoint follows a predictable pattern: https://aws-external-anthropic.<region .api.aws . Workspace operations are logged as management events in CloudTrail by default; you can enable data event logging to capture individual inference calls. What Claude Platform Has That Bedrock Does Not Here is where the feature gap starts to matter for new projects. Bedrock gives you Claude models. Claude Platform on AWS gives you the full Anthropic platform: Claude Managed Agents beta — fully hosted stateful agents, no infrastructure to manage MCP Connector beta — attach any Model Context Protocol server to your agent directly Agent Skills beta — reusable, packaged tool sets for your agents Native code execution — no Lambda wrapper required Web search and web fetch — built in, not bolted on Files API beta — persistent file storage across agent sessions Claude Console — Anthropic’s prompt development, evaluation, and improvement UI If you are building agent pipelines and you need Managed Agents or MCP Connector, Bedrock cannot give you those today. Claude Platform on AWS can. That is a meaningful architecture decision, not a minor preference. Getting Started in Under 10 Lines Setup requires four steps: subscribe in the AWS Console, link your Anthropic organization to your AWS account, create a workspace and copy the workspace ID, then install the SDK. pip install "anthropic aws " python import anthropic client = anthropic.AnthropicBedrock aws region="us-east-1", base url="https://aws-external-anthropic.us-east-1.api.aws", default headers={"anthropic-workspace-id": "YOUR WORKSPACE ID"}, message = client.messages.create model="claude-sonnet-4-6", max tokens=1024, messages= {"role": "user", "content": "Hello from Claude Platform on AWS"} , print message.content For production deployments where credentials come from an EC2 instance role or ECS task role, environment variables are cleaner: export ANTHROPIC BASE URL=https://aws-external-anthropic.us-east-1.api.aws export ANTHROPIC WORKSPACE ID=your workspace id AWS credentials come from the default provider chain — no API key needed When You Should Stay on Bedrock Claude Platform on AWS is not the right answer for everyone. Bedrock remains the better choice in specific scenarios: FedRAMP High, IL4/IL5, or HIPAA requirements — these programs require AWS as the sole data processor, which only Bedrock provides Multi-vendor model pipelines — if you need Titan, Nova, Llama, and Claude in the same inference layer, Bedrock is the unified runtime Cross-region inference profiles — Bedrock-only feature for latency optimization Guardrails and Knowledge Bases — if you have existing Bedrock investments here, migration overhead may not be worth it Claude Platform on AWS does offer an inference geo parameter — set it to us at a 1.1× price multiplier to restrict inference to US data centers. But even with that setting, Anthropic is still the data processor. If your contracts require AWS as the sole processor, Bedrock is non-negotiable. Billing: Same Price, One Invoice Token pricing is identical to the direct Anthropic API — no AWS markup. The difference is the invoice: Claude Platform usage appears on your AWS bill as a Marketplace line item, denominated in Claude Consumption Units CCUs at $0.01 per CCU. A $60 workload becomes 6,000 CCUs. For teams managing AWS budgets and cost allocation tags, consolidated billing matters more than the number on the receipt. The Bottom Line If you are starting a new Claude integration on AWS in 2026 and you do not have FedRAMP, IL4/IL5, or HIPAA constraints, Claude Platform on AWS is the stronger default. IAM authentication, CloudTrail auditing, consolidated billing, and access to Managed Agents, MCP Connector, and the Claude Console — without a separate vendor account or API key. Bedrock is the correct choice under specific compliance constraints. Outside those constraints, the feature gap is hard to ignore. Subscribe via AWS Marketplace https://aws.amazon.com/about-aws/whats-new/2026/05/claude-platform-aws/ , review the official Bedrock comparison docs https://docs.aws.amazon.com/claude-platform/latest/userguide/cpa-vs-bedrock.html , and read Anthropic’s launch post https://claude.com/blog/claude-platform-on-aws for the full architecture overview. The authentication guide https://docs.aws.amazon.com/claude-platform/latest/userguide/authentication.html covers SigV4 setup in detail if you are implementing outside the Python SDK.