Anthropic Brings Claude Platform to AWS: Native API + Managed Agents + MCP — How It Differs from Bedrock

What’s covered in this issue: In mid-May, Anthropic officially announced that Claude Platform on AWS is generally available. For the first time, AWS customers can use their own IAM credentials to call the native Claude API directly — along with Managed Agents, MCP Connector, Files API, Skills, and the full toolset. This is not an upgrade to Amazon Bedrock. It’s a parallel path that runs alongside Bedrock. This article breaks down the key differences and what they mean for your architecture.

1. The Big Event: Claude Platform Lands on AWS

Date: Mid-May 2026 Summary: Anthropic announced that Claude Platform on AWS has reached General Availability, making AWS the first cloud provider to offer the native Claude Platform experience. Sources: Anthropic / AWS joint announcement, AWS What’s New, InfoQ coverage

Capability Overview

Through Claude Platform on AWS, customers get access to:

Anthropic has explicitly committed: new features will ship simultaneously with the native Claude API — no more Bedrock-style “wait months for support” lag.

Three Pillars of AWS-Native Integration

1. Authentication: Use IAM credentials directly — no separate Anthropic API Key required
2. Billing: Consumption-based billing through AWS Marketplace, combinable with existing AWS committed spend
3. Audit: All calls flow through CloudTrail — your security and compliance team’s existing toolchain works out of the box

Available Models and Regions

Models: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5-20251001

Regions (initial 8): US East (Virginia, Ohio), US West (Oregon), Canada (Central), Brazil (São Paulo), EU (Dublin, London, Frankfurt). Asia-Pacific is not yet covered.

2. Claude Platform on AWS vs. Amazon Bedrock: What’s Actually Different?

This is the most common point of confusion. The two exist in parallel and serve fundamentally different purposes:

The one-liner: Bedrock is “Claude sold by AWS.” Claude Platform on AWS is “Claude sold by Anthropic, paid for through your AWS account.” With the former, data never leaves AWS. With the latter, data goes to Anthropic for processing. They’re not replacements for each other — they’re two paths for different compliance preferences.

3. May 19 Follow-Up: Self-Hosted Sandboxes and MCP Tunnels

Just one week after Platform on AWS launched, Anthropic announced two more capabilities at the May 19 Code with Claude event in London:

Self-Hosted Sandboxes (Public Beta)

Allows enterprises to run Claude Managed Agents’ code execution sandboxes on their own infrastructure or third-party hosts (Cloudflare, Daytona, Modal, Vercel) — keeping sensitive data within the corporate network.

⚠️ Note: Self-Hosted Sandboxes are not yet available on Claude Platform on AWS — only the native Claude API path is supported for now. Memory features are also unavailable in self-hosted sessions.

MCP Tunnels (Research Preview)

Enables Claude Managed Agents to connect to private MCP servers inside corporate networks via encrypted tunnels. This means Claude can securely access internal databases, ERPs, and custom tools without exposing those resources to the public internet.

Requires separate access approval. Targeted at financial services, healthcare, and government customers with strict data compliance requirements.

4. Practical Implications for Developers

1. Access Realities for Different Regions

The initial 8 regions are all in the Americas and Europe — no Asia-Pacific coverage yet. Even with an overseas AWS account, developers in some regions will face:

• Cross-border network latency (300–800 ms time-to-first-byte minimum)
• Cross-border data flow compliance concerns (especially for PII scenarios)
• USD billing + AWS credit card requirements

2. Teams with Existing Overseas AWS Infrastructure: Worth Migrating

If your product/service primarily serves international markets and your entire data stack already runs on AWS, Claude Platform on AWS is compelling:

• Unified billing, unified IAM, unified CloudTrail — one fewer set of accounts to manage
• Full access to Anthropic Beta capabilities (Files API, Managed Agents, MCP) — no longer blocked by Bedrock’s feature lag
• New models available day-one, no waiting for AWS to adapt

3. Strict Compliance Scenarios: Re-evaluate Your Data Boundary

The key architectural design of Platform on AWS: Anthropic is the data processor — data leaves the AWS security perimeter and enters Anthropic. For certain compliance frameworks, this is a dealbreaker.

• If compliance requires “data never leaves AWS” → stay on Bedrock
• If compliance requires “data never leaves the corporate network” → wait for Self-Hosted Sandboxes + MCP Tunnels to reach GA
• If the requirement is simply “use a US cloud provider” → Platform on AWS gets you newer features faster

4. The MCP Ecosystem Is About to Accelerate

MCP (Model Context Protocol) has been around since late 2024, and its biggest bottleneck has been “you have to write the client yourself.” Claude Platform on AWS turns MCP Connector into a managed capability, which means:

• Open-source MCP Servers can be called directly by AWS customers
• Enterprises can connect their own MCP Servers to Claude with zero client code
• Expect rapid growth in the MCP Server ecosystem over the coming months

If your team builds SaaS, consider creating an MCP Server for your product — it’s emerging as a new distribution channel in 2026.

5. Other Related Developments

6. Minimal Code Example: Try Managed Agents Beta

Here’s a minimal working example of Anthropic’s native code execution:

import anthropic

client = anthropic.Anthropic(
    api_key="sk-yourAPIkey",
    base_url="https://gw.claudeapi.com"
)

response = client.beta.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    betas=["code-execution-2025-05-22"],
    tools=[{
        "type": "code_execution_20250522",
        "name": "code_execution"
    }],
    messages=[{
        "role": "user",
        "content": "Calculate the first 50 Fibonacci numbers in Python and plot them on a logarithmic scale."
    }]
)

for block in response.content:
    print(block)

import anthropic

client = anthropic.Anthropic(
    api_key="sk-yourAPIkey",
    base_url="https://gw.claudeapi.com"
)

response = client.beta.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    betas=["code-execution-2025-05-22"],
    tools=[{
        "type": "code_execution_20250522",
        "name": "code_execution"
    }],
    messages=[{
        "role": "user",
        "content": "Calculate the first 50 Fibonacci numbers in Python and plot them on a logarithmic scale."
    }]
)

for block in response.content:
    print(block)

Calling an MCP server (e.g., GitHub MCP):

response = client.beta.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    betas=["mcp-client-2025-04-04"],
    mcp_servers=[{
        "type": "url",
        "url": "https://mcp.example.com/sse",
        "name": "github-mcp",
        "authorization_token": "ghp_xxx"
    }],
    messages=[{"role": "user", "content": "List the 5 most recent issues in the anthropic/claude-code repository."}]
)

response = client.beta.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    betas=["mcp-client-2025-04-04"],
    mcp_servers=[{
        "type": "url",
        "url": "https://mcp.example.com/sse",
        "name": "github-mcp",
        "authorization_token": "ghp_xxx"
    }],
    messages=[{"role": "user", "content": "List the 5 most recent issues in the anthropic/claude-code repository."}]
)

Note: Managed Agents and MCP Connector are both in Beta — you must include the corresponding beta header in your calls. The API surface may change across versions; refer to the Anthropic official release notes.

Summary

Claude Platform on AWS is Anthropic’s most significant product architecture move of 2026 — delivering the “native Claude platform” into enterprise IT stacks with AWS accounts as the entry point, creating a clear choice alongside Bedrock between “data boundary” and “feature completeness.”

Near-term practical impact for developers:

• Teams serving international markets: Evaluate migrating to Platform on AWS for day-one Beta access + unified billing
• Teams focused on domestic markets: Continue using the Anthropic native API path (self-hosted proxy or relay services like claudeapi.com) — latency and cost are typically better
• MCP builders: The launch of Managed MCP Connector opens a new distribution opportunity

The next milestones to watch: Anthropic’s IPO timeline (media reports suggest October 2026) and when Self-Hosted Sandboxes gain AWS Platform support.