Using Claude Desktop / Code with Bedrock | Key Setup and Pricing Points

What It Means to Run Claude Desktop on Bedrock — The Third-Party Inference Mode Officially Enabled in 2026

Claude Desktop is Anthropic's official desktop application for Mac and Windows. By default, it connects to Anthropic-hosted inference infrastructure and requires signing in with a Pro, Max, Team, or Enterprise plan.

In 2026, AWS and Anthropic officially launched a public beta of "third-party inference (3P inference)" that lets Claude Desktop direct its inference to Amazon Bedrock. All prompts, files, and tool call results entered by users are processed through Bedrock, with nothing passing through Anthropic's inference servers (Source: From developer desks to the whole organization: Running Claude Cowork in Amazon Bedrock).

The same mechanism allows Claude Code (the CLI) to run on a Bedrock backend as well, making it possible to shift both the desktop app and the development terminal experience entirely to your AWS account. For companies that have avoided using Anthropic directly due to data governance requirements, or for developers who hold AWS Credits, this has become a practical alternative to the Anthropic API.

Requirements — Model IDs, Inference Profiles, and Authentication

To enable the Bedrock backend, you need the following in place beforehand:

1. AWS account and IAM permissions — An IAM role or user that includes bedrock:InvokeModel and related permissions.
2. Bedrock model access enabled — The Claude models you want to use (Opus 4.x, Sonnet 4.6, Haiku 4.5, etc.) must have been requested and approved in the AWS Console. Note that Claude model availability on Bedrock may lag behind direct Anthropic contracts, so the latest generation of Opus may not yet be available. Always verify that the inference profile ID you plan to use is actually listed in the AWS Console model catalog before proceeding — using a copied ID that is not yet available will cause InvokeModel to fail with an unknown model error.
3. Inference profile ID — You must specify a region-prefixed inference profile ID such as us.anthropic.claude-opus-4-7-v1, not a bare model ID like anthropic.claude-opus-4-7-v1 (Source: Running Claude Code and Claude Desktop on Amazon Bedrock — DEV Community).
4. Authentication method — Choose either AWS IAM (~/.aws/credentials or environment variables) or an Amazon Bedrock API key.
5. Supported region — A region where Claude models are available on Bedrock (us-east-1, us-west-2, ap-northeast-1, etc.).

An inference profile is Bedrock's routing unit for distributing inference requests across multiple regions — a prefix like us. indicates a cross-region (US-based) profile. Without this prefix, InvokeModel cannot determine which region to use and will fail, which is why the us.-prefixed inference profile ID is required. Inference profiles come in three types — Regional, Cross-Region, and Global — and if you have no data residency requirements, Cross-Region or Global profiles offer a good balance of availability and cost. You can also pin to the Tokyo region if needed.

Claude Desktop Setup — Developer Mode and UI Configuration

When configuring Bedrock in Claude Desktop, the official recommendation is to use the Setup UI rather than editing JSON directly to avoid configuration errors.

Step 1: Enable Developer Mode

Launch the Claude Desktop app and select Help → Troubleshooting → Enable Developer mode from the menu bar. A Developer item will then appear in the main menu.

Step 2: Configure Third-Party Inference

Open Developer → Configure third-party inference and enter the following:

• Inference provider: Bedrock
• AWS region: e.g., us-east-1 or whichever region you want to use
• AWS profile name: typically default (the profile name in ~/.aws/credentials)
• Model list: an inference profile ID such as us.anthropic.claude-opus-4-7-v1

Step 3: Verify After Applying

Save the settings and restart the app. Even while signed in, all model calls will now be directed to Bedrock. Your Anthropic Pro/Max seat will not be consumed. For organizational rollouts, you can push the same settings via MDM (Jamf, Microsoft Intune, or Group Policy) to apply them to all employees at once.

Claude Code Setup — settings.json and Environment Variables

For Claude Code (CLI), add the following to ~/.claude/settings.json:

{
  "model": "us.anthropic.claude-opus-4-7-v1",
  "env": {
    "CLAUDE_CODE_USE_BEDROCK": "1",
    "AWS_REGION": "us-east-1"
  }
}

An important point: Claude Code does not automatically read the region setting from ~/.aws/config, so you must explicitly specify AWS_REGION in settings.json. If omitted, the bedrock:InvokeModel call will fail with no region specified.

After configuring, launch claude in the terminal and type /status. If it shows API provider: Amazon Bedrock, the switch was successful. Make sure your IAM credentials are active in advance via the AWS CLI (aws configure or aws sso login).

Unavailable Features — Chat Tab, Computer Use, and Skills Marketplace

In Bedrock inference mode, some features that depend on Anthropic-hosted inference are disabled. Be sure to assess the impact before deploying.

• Chat tab — The chat UI equivalent to claude.ai runs on Anthropic's side and is not available via Bedrock. Only the Cowork and Code tabs function.
• Computer Use — The browser automation feature using mouse and keyboard input depends on Anthropic-hosted inference and is not available with Bedrock alone (Bedrock does offer partial access through a separate API path: Use a computer use tool to complete an Amazon Bedrock model response).
• Skills Marketplace — Distribution of Skills from the official marketplace also depends on Anthropic's infrastructure and is not supported.

On the other hand, Projects, Artifacts, Memory, File upload, Remote connectors, MCP servers, and some Plugins work via Bedrock as well. For Claude Code integration or everyday document generation and research tasks, the functional difference is minimal.

Pricing and Data Handling — AWS Pay-as-You-Go and No Training Use

Billing is consolidated into pay-as-you-go charges under your existing AWS contract, with no Anthropic seat license fees. Even a 10,000-person organization only needs to issue IAM roles to get started. Charges appear through the normal AWS billing channel and can be visualized in Cost Explorer or CUR (Cost and Usage Report).

Per-token pricing follows the standard Bedrock rates for Claude. As a baseline, Sonnet 4.6 is $3 input / $15 output, Haiku 4.5 is $1 input / $5 output, and Opus 4.7 is $5 input / $25 output (all per 1M tokens). Prompt Caching and Batch API discounts are also available on the Bedrock side. See the Bedrock Claude Pricing article for a detailed breakdown.

On the data handling side, Amazon Bedrock does not store prompts, files, tool calls, or model responses during inference, and does not use them to train foundation models. This policy is explicitly stated in the official AWS blog (Source: AWS Machine Learning Blog). The design makes it suitable for adoption in finance, healthcare, and public sector environments with data residency requirements.

How This Article Fits with Related Content — Cowork, Pricing, and More

This article focuses specifically on setting up Claude Desktop / Code via Bedrock for individual developers and small teams. The following related articles complement this one for specific use cases:

• Bedrock Claude Cowork | Key Points for Organizational Deployment and IAM Cost Allocation — Covers MDM distribution for company-wide rollouts, IAM principal-based cost allocation, CloudTrail auditing, and guardrail design. Aimed at IT and platform teams.
• Bedrock Claude Pricing | Per-Model Rates for Opus, Sonnet, and Haiku with Cost Reduction Tips — Latest per-model pricing, differences between global and regional profiles, and examples of Prompt Caching and Batch API usage.
• Using Claude on AWS | How to Use Amazon Bedrock and Its Pricing — Criteria for choosing between Bedrock and Claude Platform on AWS.
• Claude Desktop App | How to Use Its 5 Key Features and Common Pitfalls — A full feature overview of Claude Desktop when used with a direct Anthropic contract.

A recommended reading path: individual developers should read this article then the pricing article; for organizational rollouts, read this article → the Cowork article → the pricing article; and for top-level AWS selection decisions, start with the AWS overview article. Whether to switch from a direct Anthropic contract to Bedrock ultimately comes down to your usage of the three features — Chat tab, Computer Use, and Skills Marketplace — balanced against your AWS Credit balance and data residency requirements.