Claude/Desktop/Q/Is it possible to use AWS Bedrock to power Claude Desktop?
Is it possible to use AWS/Bedrock to power Claude/Desktop?
- AI/Response
- Short answer (as of 2026-05-28): Yes. Claude/Desktop can route model inference through Amazon Bedrock in your AWS account via third-party inference configuration (the same Cowork / 3P path used for enterprise Bedrock deployments). This is distinct from the Claude/Code CLI’s
CLAUDE_CODE_USE_BEDROCK=1 env-var flow, but serves the same goal: use Bedrock instead of Anthropic-hosted inference endpoints. - How to enable (evaluation / single machine):
- Enterprise rollout: IT can push the same keys (
inferenceBedrockRegion, inferenceBedrockProfile, inferenceBedrockBearerToken, SSO fields, inferenceModels, optional gateway URL) via MDM (Jamf, Intune, Group Policy). AWS describes Claude Cowork in Amazon Bedrock as inference in your account/regions with optional LLM gateway routing (AWS ML blog, Apr 2026). - What Bedrock covers vs. not:
- Through Bedrock: model calls to
bedrock-runtime in configured regions; billing/audit via AWS (IAM, CloudTrail, VPC endpoints as you design). - Still Anthropic-side or not Bedrock-backed today: some product surfaces called out in AWS/Anthropic 3P docs as not available through Bedrock inference (e.g. certain hosted capabilities, depending on release). MCP connectors can still reach endpoints you approve; aggregate telemetry to Anthropic can be reduced/disabled per 3P policy docs.
- Compare Claude/Code: CLI uses env vars such as
CLAUDE_CODE_USE_BEDROCK and Bedrock auth refresh settings (Claude/Code/Bedrock). Desktop uses the in-app 3P inference JSON/managed config instead.
- Prerequisites: Bedrock model access in the chosen region (
bedrock:InvokeModel, streaming invoke as needed), valid inference profile IDs in inferenceModels for profile/SSO auth, and a Desktop build that includes 3P Bedrock (feature landed in 2026; tracked publicly from anthropics/claude-code#32668).