sagemaker-ai Claude Code Plugin
Claude Code PluginClaude CodeDevelopmentAI Agents & AI App DevelopmentCloud, Deployment & CI/CDThe explanation below is AI-generated. Please verify it against the sources.
This plugin brings AWS SageMaker AI expertise directly into coding assistants such as Claude Code and Cursor. It also bundles an MCP server that retrieves AWS documentation and standard operating procedures. Using it requires an AWS account with an IAM role holding SageMaker (and, for some paths, Bedrock and Lambda) permissions, and it can be run either from a remote connection to SageMaker Studio Spaces or from a locally configured compute environment.
About the service
Amazon SageMaker AI, as described on its product page, is AWS's platform for building, training, and deploying machine learning and foundation models. It sits within the broader next-generation SageMaker offering, which also provides SageMaker Unified Studio, a data/AI catalog, and a lakehouse architecture spanning Amazon S3 and Amazon Redshift so that model development, generative AI application building, and SQL analytics can happen in one governed environment. The plugin's deployment and evaluation skills can also target Amazon Bedrock as an alternative to SageMaker endpoints.
What you can do with sagemaker-ai
- Walk through a guided plan for fine-tuning a foundation model, from use-case definition to deployment
- Validate dataset quality and convert datasets into SageMaker-compatible training formats
- Choose fine-tuning techniques such as SFT, DPO, or RLVR and select a base model
- Design model evaluations, including LLM-as-a-judge comparisons and benchmark selection
- Configure and launch a deployment endpoint on SageMaker AI or Amazon Bedrock
- Run remote commands and transfer files on HyperPod cluster nodes via AWS Systems Manager
- Diagnose HyperPod issues such as cluster creation failures, EFA health, NCCL/AllReduce errors, node hardware faults, performance bottlenecks, and Slurm scheduling problems
- Look up AWS documentation and SOPs through the bundled aws-mcp server