View a markdown version of this page

Installing Amazon SageMaker AI skills - Amazon SageMaker AI
Agent SkillsMCP ServersInstall Skills with npx skills
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Installing Amazon SageMaker AI skills

The Agent Toolkit for Amazon gives AI coding agents the tools, knowledge, and science-based best practices they need to work with Amazon services. It works with the coding agents you already use, including Claude Code, Codex, Cursor, and Kiro. The toolkit bundles the Amazon MCP Server configuration and a curated set of agent skills in a single install, so your agent can discover and apply Amazon best practices automatically without you having to know which skill to invoke.

The toolkit includes a broad set of core skills that span many Amazon services and common workflows, such as Amazon Bedrock and Amazon Elastic Compute Cloud, in addition to the Amazon AI/ML skill covered on this page. Your agent loads only the skills relevant to the task at hand.

The Amazon AI/ML skill (aws-ai-ml) brings deep Amazon AI/ML expertise into your coding assistant and covers Amazon SageMaker AI. It supports the full model customization lifecycle from planning through production, and it currently assists with the following capability areas:

  • Model selection — Guided selection of a base model from Amazon SageMaker AI Hub, matching model family and size to your use case requirements, for either fine-tuning or off-the-shelf deployment.

  • Model deployment — Deployment configuration and endpoint setup on Amazon SageMaker AI or Amazon Bedrock, covering the Nova and open-source (OSS) deployment pathways.

  • Model fine-tuning — End-to-end guided workflows for fine-tuning foundation models, from use case definition through data preparation, training, evaluation, and deployment on Amazon SageMaker AI. Supports both serverful and serverless paths.

  • Model evaluation — Evaluation design, benchmark selection, LLM-as-a-judge and custom scorers, and side-by-side model comparison to measure quality before and after customization.

  • Inference optimization — Benchmarking and tuning of real-time endpoints to meet performance, cost, latency, and throughput goals, including instance recommendations.

  • MLflow — Set up, update, or delete a Amazon SageMaker AI Managed MLflow app to track experiments, parameters, and metrics across the customization lifecycle.

Agent Skills

The following skill is installed by the plugin:

Amazon SageMaker AI agent skills
Skill Description Documentation
aws-ai-ml Selects, deploys, and customizes AI models on Amazon SageMaker AI. Covers the full lifecycle from planning through production: model selection, dataset preparation, fine-tuning (SFT, DPO, RLVR, RLAIF), evaluation, deployment to Amazon SageMaker AI endpoints or Amazon Bedrock, inference optimization, endpoint diagnostics, and Amazon SageMaker AI Managed MLflow. SKILL.md

MCP Servers

Amazon SageMaker AI Skills requires the Amazon SageMaker AI MCP server. Add the contents of the .mcp.json file to your platform's MCP configuration file:

  • Claude Code: Run claude mcp add --transport stdio aws-mcp -- uvx mcp-proxy-for-aws@latest https://aws-mcp.us-east-1.api.aws/mcp or manually add to User/Project/Local location as needed (Claude Code Docs: What uses scopes).

  • Cursor: .cursor/mcp.json

  • Kiro: .kiro/settings/mcp.json

Install Skills with npx skills

You may use the Skills CLI (from Vercel Labs) to install the skills into your platform:

  • Claude Code:

    npx skills add aws/agent-toolkit-for-aws/skills/core-skills/aws-ai-ml --all --agent claude-code --copy

  • Cursor:

    npx skills add aws/agent-toolkit-for-aws/skills/core-skills/aws-ai-ml --all --agent cursor --copy

  • Kiro:

    npx skills add aws/agent-toolkit-for-aws/skills/core-skills/aws-ai-ml --all --agent kiro-cli --copy

If you have configured other agents, substitute your agent name for <agent> and use:

npx skills add aws/agent-toolkit-for-aws/skills/core-skills/aws-ai-ml --all --agent <agent>

Alternatively, install the full aws-core plugin, which bundles the Amazon MCP Server configuration and the curated core skill set in a single install:

  • Claude Code:

    /plugin install aws-core@claude-plugins-official
/reload-plugins

  • Codex:

    codex plugin marketplace add aws/agent-toolkit-for-aws

    Then launch Codex, run /plugins, and install the aws-core plugin.

  • Amazon CLI (version 2.35.0 or later):

    Run the interactive setup wizard, which detects your installed AI coding agents, installs default Amazon skills, and configures the Amazon MCP Server connection in a single command:

    aws configure agent-toolkit