AWS ships AgentCore CLI 1.0 for building and deploying Bedrock agents in TypeScript

AWS releases AgentCore CLI 1.0

AWS made version 1.0 of the AgentCore CLI generally available on October 28, 2026, according to a hands-on walkthrough published on dev.to. The CLI is the working entry point to Amazon Bedrock AgentCore, the AWS platform for operating AI agents in production. It generates agent projects, runs them locally for testing, and ships them to AWS. Deployment relies on AWS CDK under the hood, but the tooling hides that from the user.

The npm package @aws/agentcore had existed in preview since February 2026 as a series of 0.x releases. For 1.0, the dev.to author reports, AWS rebuilt the CLI: it starts faster, follows one uniform command structure and covers more resource types. The author tested everything against version 1.0.1 and estimates that a TypeScript agent with custom tools needs roughly five commands and two files to go from an empty directory to a live call in AWS.

Four concepts to know

AgentCore splits into building blocks that can be used alone or together, and the CLI manages them. The walkthrough boils the platform down to four terms. The Runtime is the managed environment for agent code, giving every session its own isolated microVM. The Harness is an agent described purely through configuration — model, prompt and tools — where AgentCore runs the agent loop and no code is written. Memory stores conversation histories and facts derived from them, so an agent can recall things across sessions. The Gateway exposes APIs and Lambda functions as MCP tools for agents to call.

What is new in the CLI

Commands are grouped into three sections: Project (create, dev, deploy, invoke), Resources (runtime, harness, gateway, memory) and Settings. Existing resources can be addressed directly by ID or ARN without first being attached to a project. Global options for region, JSON output and debugging behave identically across every command, which the author notes makes CI scripts far more pleasant. Running the bare command with no arguments launches an interactive terminal UI with menus and wizards.

The TypeScript path

The practical requirements are modest: Node.js 20.12 or newer, valid AWS credentials, IAM permissions for AgentCore and the CDK bootstrap roles, and access to a Claude model in Bedrock. Docker is not needed, since the TypeScript template is packaged as a ZIP archive.

Project creation uses a template flag; the default wizard produces a harness, so developers wanting their own code pick a template such as agent-typescript-strands, which builds on Strands Agents, an open-source AWS framework. A Vercel AI SDK variant and several Python templates also exist. The generated project centres on an agentcore.json file that lists which resources get deployed and is translated into a CDK stack, alongside a generated CDK project where developers can add their own AWS resources, with a DynamoDB example included. Tools are functions the model may call, with Strands validating arguments against a Zod schema.

One deliberate simplification in the walkthrough: the Strands template automatically attaches a Memory with four strategies — facts, preferences, summaries and episodes — which the author removed for the first run because it costs extra, slows deployment and obscures what the agent itself does. Memory can be added back once an agent genuinely needs to remember things.

Rough edges to watch

Projects built on 0.x versions must be converted with agentcore migrate, and parts of the online documentation still describe old command names, such as import instead of migrate. The author recommends relying on per-command help output for the first weeks. An older Python toolkit, bedrock-agentcore-starter-toolkit, also installs a binary called agentcore and should be uninstalled to avoid clashes. The CLI collects anonymous usage data by default, which can be switched off through a config setting. Developers who start with a code-free harness are not locked in: it can be exported to code later if the configuration-only approach becomes limiting.

Why it matters

Plenty of agent prototypes never leave a developer's machine, and the dev.to author argues the reason is rarely the model — it is the unglamorous plumbing of runtime, permissions and logging in between. AgentCore CLI 1.0 takes direct aim at that gap for AWS shops, turning deployment into a small number of commands rather than bespoke infrastructure work. The uniform command structure and consistent output flags also make the tool plausible for CI pipelines rather than one-off local use. The short-term costs are real — a forced migration from 0.x and documentation that lags behind the release — but for teams standardising on Bedrock, the CLI lowers the barrier between a working prototype and an agent actually running in production.