AWS Agent Toolkit

Official AWS-supported bundle of plugins, skills, rules, and MCP server configuration for agents building, deploying, and managing AWS applications.

What it is

aws/agent-toolkit-for-aws gives coding agents AWS-specific skills and a managed AWS MCP Server route. The README names Claude Code, Codex, Cursor, and Kiro as supported agent surfaces and includes install paths for AWS plugins and skills. Source: GitHub README, 2026-06-26

The toolkit currently includes:

Surface What it covers
aws-core plugin service selection, CDK/CloudFormation, serverless, containers, storage, observability, billing, SDK usage, deployment
aws-agents plugin AI agents on AWS, including Amazon Bedrock and AgentCore
aws-data-analytics plugin S3 Tables, AWS Glue, Athena, data lake, analytics, and ETL workflows
aws-agents-for-devsecops plugin incident investigation, code review, UAT readiness, vulnerability scanning, penetration-test support
AWS MCP Server AWS API coverage, sandboxed script execution, AWS documentation search, CloudWatch metrics, IAM condition keys, CloudTrail audit logging

Source: GitHub README, 2026-06-26

Routing

Use AWS Agent Toolkit when an agent is expected to reason about AWS architecture, docs, service selection, CDK/CloudFormation, Bedrock/AgentCore, data/analytics, or AWS security workflows.

Use the local aws CLI when you need normal account-specific reads/writes and already know the exact AWS API call. Use Floci for laptop/CI AWS-shaped tests when real cloud mutation is unnecessary. For any mutation, confirm account, region, and environment first.

Guardrails

This page does not mean "let agents mutate AWS freely." The useful parts are AWS-supported documentation search, skills, rules, audited plugin bundles, and MCP logging/audit controls. Writes still follow the repo's cloud mutation rule: identify target account, read/list first, dry-run or plan when possible, and ask before production or cost-bearing changes.


Timeline