The AWS Command Line Interface (CLI) now supports the Agent Toolkit for AWS
One command now installs and governs AI coding agents across AWS—with live API access, curated skills, and enterprise audit controls built in.
View original announcement →Visual Summary
graph TD
A{{AWS CLI Agent Toolkit}}:::announced
B((Developer)):::external
C([Interactive Wizard]):::feature
D([Skill Management]):::feature
E(AWS MCP Server):::compute
F([40+ Agent Skills]):::feature
G(Coding Agents):::compute
H(AWS APIs 15000+):::compute
I(IAM & CloudWatch):::compute
B ==>|"aws configure agent-toolkit"| A
A -->|"launches"| C
A -->|"provides"| D
C -->|"detects & configures"| G
A -->|"installs"| E
A -->|"installs"| F
E -->|"secure interface"| H
F -.->|"injects knowledge"| G
E -->|"auditable access"| G
E -->|"monitored by"| I
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
AWS has integrated the Agent Toolkit for AWS directly into the AWS CLI, enabling developers to install and configure AI coding agent tooling across multiple agents—including Kiro, Claude Code, Codex, and Cursor—using a single interactive wizard command (aws configure agent-toolkit). The Agent Toolkit provides coding agents with a secure MCP Server interface to 15,000+ AWS APIs, 40+ curated skills covering domains like storage, networking, and analytics, and three bundled plugins. Developers can also use the new aws agent-toolkit command group to search, install, update, and remove individual skills from the command line.
How It Works
- Interactive setup wizard: Running
aws configure agent-toolkitlaunches a wizard that automatically detects installed coding agents by scanning for their configuration directories (e.g.,~/.kiro) and walks users through installing the AWS MCP Server and default skills across all detected agents in one step. - AWS MCP Server: Acts as a secure, auditable intermediary between coding agents and AWS, exposing 15,000+ AWS APIs through the Model Context Protocol (MCP) with IAM-based access controls, CloudWatch monitoring, and enterprise guardrails that distinguish agent actions from human actions.
- Agent Skills: 40+ curated skill packages deliver up-to-date documentation, best practices, and tested multi-step procedures (e.g., configuring data pipelines, creating production-ready VPCs) directly to the agent's context, reducing trial-and-error.
- Skill management commands: The
aws agent-toolkitcommand group supportsadd-skill,update,remove,list, andsearchoperations; the--agentflag targets a specific tool, and--skill-versionpins a specific version. - Global skill installation: Skills are installed globally into agent configuration directories rather than per-project, so all projects using that agent benefit automatically.
- Plugin bundles: Three plugins bundle the MCP Server with curated skill sets into a single install for common use-case patterns, simplifying onboarding further.
- Prerequisites: AWS CLI version 2.35.0 or later is required, along with at least one supported coding agent installed on the system.
Why It's Important
- Reduces agent hallucination and outdated knowledge: Foundation models have training cutoffs, so agents often lack awareness of newer AWS services (e.g., S3 Tables, Aurora DSQL, Bedrock AgentCore); the toolkit injects real-time, current AWS documentation directly into agent context.
- Improves production readiness: Instead of scaffolding generic code, agents equipped with skills can produce infrastructure that follows AWS Well-Architected best practices, correct security configurations, and cost-efficient resource choices from the start.
- Cuts token waste and cost: Tested procedures and accurate API knowledge mean agents get tasks right on the first attempt, dramatically reducing the token-burning trial-and-error loops that inflate costs.
- Enables enterprise governance: The MCP Server layer introduces auditability and IAM-based controls over agent-executed AWS actions—something impossible when agents invoke the CLI directly from a terminal.
- Dramatically simplifies multi-agent setup: Previously, configuring MCP servers and skills across multiple coding agents required manual, per-agent configuration; the CLI wizard collapses this into a single command.
- Extensible and maintainable: The skill management CLI makes it easy to keep agent knowledge current as AWS evolves, treating skills as versioned, updatable packages rather than static prompts.
How It's Different
- CLI-native experience vs. manual JSON editing: Competing or prior approaches required developers to manually edit MCP configuration files for each agent; the AWS CLI wizard automates detection and configuration across all installed agents simultaneously.
- Auditable agent interface vs. raw CLI execution: When agents run AWS CLI commands directly in a terminal, there is no differentiation from human actions and no audit trail; the MCP Server provides CloudWatch metrics, IAM controls, and enterprise guardrails as a dedicated agent interface.
- Curated, versioned skills vs. static system prompts: Rather than embedding AWS knowledge in a one-time system prompt, skills are versioned packages that can be updated independently as services evolve, ensuring agents always have current guidance.
- 15,000+ API coverage vs. agent general knowledge: General-purpose coding agents rely on training data that may be months or years stale; the MCP Server provides live access to the full AWS API surface, including recently launched services.
- Cross-agent portability: The same toolkit and skills work across Kiro, Claude Code, Codex, Cursor, and any MCP-compatible agent, avoiding vendor lock-in to a single coding assistant.
- Skill granularity and composability: Customers can install only the skills relevant to their workload (e.g., serverless, networking) rather than loading all context at once, keeping agent context windows focused and efficient.
When to Prefer It
- Teams using multiple coding agents: When developers on the same team use different AI assistants (e.g., some on Cursor, others on Claude Code), the CLI wizard ensures consistent AWS tooling and guardrails across all of them in one pass.
- Enterprise environments requiring governance: When organizational policy requires audit logs of agent-executed AWS actions or IAM-based restrictions on what agents can do versus humans, the MCP Server layer is essential.
- Working with recently launched AWS services: When a project involves services launched after a model's training cutoff (e.g., Aurora DSQL, Bedrock AgentCore), skills provide agents with accurate, up-to-date documentation they would otherwise lack.
- Production infrastructure provisioning: When the goal is deploying production-grade infrastructure (not just prototypes), skills guide agents through Well-Architected patterns, security configurations, and cost-optimized choices.
- Complex multi-step AWS workflows: When tasks involve intricate sequences like configuring a data pipeline, setting up a VPC with proper routing, or wiring Lambda to API Gateway, curated skills provide tested step-by-step procedures that reduce errors.
- Cost-sensitive agentic workloads: When minimizing LLM token consumption matters, equipping agents with precise AWS knowledge upfront eliminates expensive retry loops caused by incorrect API calls or outdated service knowledge.
- Developers new to a specific AWS domain: When a developer or agent needs to work in an unfamiliar AWS domain (e.g., analytics, networking), domain-specific skills provide expert guidance without requiring deep prior knowledge.
Availability
- General Availability: The AWS CLI integration (
aws configure agent-toolkitandaws agent-toolkitcommands) is generally available as of June 5, 2026. - CLI version requirement: AWS CLI version 2.35.0 or later is required; earlier versions do not support the new command groups.
- AWS MCP Server regions: The AWS MCP Server is currently available only in US East (N. Virginia) and Europe (Frankfurt); other regions are not yet supported.
- Supported coding agents: Kiro, Claude Code, Codex, and Cursor are explicitly supported; any other MCP-compatible coding agent should also work.
- Pricing: No separate pricing is mentioned for the Agent Toolkit itself; standard AWS service charges apply for any AWS resources provisioned or APIs called through the toolkit, and CloudWatch charges apply for monitoring data.
- Skill availability: 40+ skills are available at launch, searchable and installable via the CLI; skills are installed globally per agent, not scoped to individual projects.
- Open source: The toolkit is available on GitHub, indicating an open-source component alongside the managed MCP Server service.
