# awslabs/agentcore-samples: a tour of the AgentCore sample repository and its CLI migration

> The samples repository for Amazon Bedrock AgentCore collects runnable examples of Runtime, Gateway, Memory, Identity and Evaluation. It is also mid-migration from the Starter Toolkit to the AgentCore CLI, which is the first thing to understand before you copy any code.

**awslabs/agentcore-samples** — Amazon Bedrock Agentcore accelerates AI agents into production with the scale, reliability, and security, critical to real-world deployment.

- Repository: https://github.com/awslabs/agentcore-samples
- Website: https://aws.amazon.com/bedrock/agentcore/
- Stars: 3,415 · Forks: 1,340
- Language: Python
- License: Apache-2.0
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/awslabs-agentcore-samples

## What the repository actually contains, and who it is for

This is a samples and tutorials repository, not a library you install as a dependency. The README describes Amazon Bedrock AgentCore as framework-agnostic and model-agnostic infrastructure for deploying and operating agents, and the repository exists to show how to use it. The audience is engineers who have already decided to run agents on AgentCore and now need concrete code: how a Runtime entrypoint is shaped, how a Gateway turns an API into an MCP-compatible tool, how Memory is wired in.

The top-level layout is the map. getting-started/ holds Python and TypeScript samples for a first agent. features/ holds deep dives on Runtime, Gateway, Identity, Memory, Tools, Observability, Evaluation and Policy. end-to-end/ holds complete applications that combine several capabilities, and the README says each includes deployment instructions, architecture diagrams and testing guides. integrations/ covers identity providers, observability platforms, data platforms and UX examples. infrastructure-as-code/ carries CloudFormation, AWS CDK and Terraform templates, and blueprints/ holds full-stack reference applications.

One structural detail matters more than the rest: the README states the repository is transitioning from the Bedrock AgentCore Starter Toolkit to the AgentCore CLI, that samples still depending on the Starter Toolkit live in legacy/, and that MIGRATION.md carries the old-path to new-path mapping. Any example you copy without checking that file may be written against a tool the README says is no longer supported.

## How the samples are organised around AgentCore capabilities

The repository does not implement agent infrastructure. It demonstrates the AgentCore services, and the README's own framing is that AgentCore removes the undifferentiated heavy lifting of building and managing specialized agent infrastructure so you can bring a framework and a model without rewriting code. The samples are therefore thin by design: each one configures a capability and shows the call pattern.

features/ is the clearest expression of that. Runtime is described as a secure, serverless runtime for deploying agents and tools at scale. Gateway converts APIs, Lambda functions and services into MCP-compatible tools. Identity handles agent identity and access management across AWS and third-party apps. Memory is managed memory infrastructure for personalized agent experiences. Tools covers the built-in Code Interpreter, Browser Tool and Web Search Tool. Observability uses OpenTelemetry to trace, debug and monitor agents. Evaluation offers built-in and custom evaluators for on-demand and online evaluation, and Policy applies fine-grained access control with Cedar policies.

That list is also the honest limit of what the repository tells you. Each bullet links to the AWS developer guide rather than explaining the service in the repository itself, so the samples are a companion to that documentation, not a replacement for it. If you want to understand what AgentCore Memory does internally, this repository will not tell you. It will show you a sample that calls it.

## Installing the samples and running a first example

The repository is cloned, not installed. The README points to the AgentCore CLI (@aws/agentcore) as the recommended way to create, develop and deploy agents, and describes it as supporting Strands, LangGraph, LangChain, Google ADK, OpenAI Agents and bring-your-own frameworks, with local development including hot reload, built-in evaluations and gateway support. The getting-started/ directory is where the first agent lives, split into python/ and typescript/.

The top-level requirements.txt lists the packages the samples expect, including strands-agents, bedrock-agentcore, bedrock-agentcore-starter-toolkit, langchain[aws], langgraph, mcp>=1.9.0, boto3 and jupyterlab. Because the repository ships a uv.lock alongside it, uv is the dependency tool the lock file implies, and uv appears in requirements.txt itself:

```
uv
```

Note what that requirements file also tells you: bedrock-agentcore-starter-toolkit is still listed even though the README says the Starter Toolkit CLI is no longer supported and the AgentCore CLI is the recommended path. Treat the requirements file as describing the repository as a whole, not as a statement about which CLI a given sample uses. Before running anything, open the sample's own directory and read its README, then check MIGRATION.md for where that sample sits in the old-path to new-path mapping. The README also links a video walkthrough for building a first production-ready agent, which is the closest thing to a guided first run the repository offers.

## The migration is the biggest practical obstacle

A samples repository is only as useful as the code inside it is current, and this one is openly in transition. The README says samples that still depend on the Starter Toolkit are in legacy/ and will be updated over the coming weeks, and the workshops section is marked deprecated legacy code with an explicit recommendation to use the AgentCore CLI instead. That is unusually candid, and it is also a warning.

In practice this means two samples that look equally authoritative can target different toolchains. If you copy a Starter Toolkit pattern into a project built on the AgentCore CLI, you may be reproducing an approach the maintainers are moving away from. The repository does provide the remedy: MIGRATION.md is described as containing the full old-path to new-path mapping. Checking it before copying is cheap; discovering the mismatch after you have wired a deployment is not.

The second limitation is scope. This is a samples repository, so it optimises for readability over production hardening. The pyproject.toml makes that explicit in its Ruff configuration, which relaxes E402, E722, F401, F811 and F841, with comments explaining that tutorial scripts configure the environment before imports, use bare except clauses for readability, and assign variables for clarity even when unused. Those are reasonable choices for teaching code. They are also exactly the patterns you would not want copied into a service. Read the samples for the call patterns, then write your own error handling.

## How it compares with building directly on the SDK

The obvious alternative is the Python SDK at aws/bedrock-agentcore-sdk-python, which the README links alongside the documentation and the AgentCore CLI. The difference is one of abstraction level rather than capability. The SDK is the interface you program against. This repository is a set of worked examples showing how that interface is used across Runtime, Gateway, Memory, Identity, Tools, Observability, Evaluation and Policy, plus end-to-end applications and blueprints that combine them.

Choosing between them is not really a choice. You will use the SDK either way. What the samples add is the shape of a working configuration and the integration patterns for third parties: Okta, Entra and Cognito under identity-providers/, Grafana, Datadog and Dynatrace under observability/, and Streamlit and AG-UI under ux-examples/. Those integration directories are where the repository earns its place, because they answer questions the SDK reference will not, such as how an external identity provider is expected to connect to AgentCore Identity.

A second alternative is to skip samples entirely and start from the AgentCore CLI's own scaffolding. The README presents the CLI as the fastest way to create, develop and deploy agents, with local development, hot reload and built-in evaluations. If your goal is a new agent rather than an understanding of AgentCore, generating a project with the CLI and consulting these samples when you hit a specific capability is a shorter path than reading the repository front to back.

## Maintenance, licensing and what to check before you commit

The repository is not archived, and the last push was on 2026-09-09, which is recent. The README also carries a commit-activity badge, though activity alone says nothing about whether a specific sample has been migrated. The README's own statement that legacy samples will be updated over the coming weeks is the more useful signal: expect churn in the legacy paths and relative stability in getting-started/ and features/.

The licence is Apache-2.0, and the repository ships a NOTICE file alongside LICENSE, which is the normal arrangement for Apache-2.0 projects that include attribution requirements. Apache-2.0 permits commercial use and modification and includes a patent grant; it also requires that you preserve copyright and attribution notices. That is a description of the licence text, not legal advice, and if you plan to redistribute sample code inside a product, the NOTICE file is the thing to read first.

Upgrade cost is concentrated in the CLI transition. Because the README frames the move from the Starter Toolkit to the AgentCore CLI as an in-progress migration with a mapping document, the maintenance burden for anyone tracking this repository is keeping their own code aligned with whichever CLI the maintainers settle on. There are no retrieved releases for this repository, so there is no versioned changelog to pin against. You are tracking a branch.

## Conclusion

Adopt these samples if you are building on Bedrock AgentCore and want working reference code for Runtime, Gateway, Memory or Identity, and if you are willing to check every example against MIGRATION.md before relying on it. Do not adopt the repository as a substitute for the AgentCore service documentation, and do not treat the legacy Starter Toolkit paths as current, because the README states that the Starter Toolkit CLI is no longer supported. Verify first whether the sample you want lives under getting-started/ or under legacy/, and confirm which CLI the sample's own README names, since the top-level README says samples depending on the Starter Toolkit are still being updated over the coming weeks.

## FAQ

### What is Amazon Bedrock AgentCore?

According to the README, it is framework-agnostic and model-agnostic infrastructure for deploying and operating AI agents securely and at scale, so you can bring a framework such as Strands, CrewAI, LangGraph or LlamaIndex and any LLM without rewriting code.

### What are the key features of AgentCore?

The repository's features/ directory lists Runtime for serverless agent deployment, Gateway for turning APIs and Lambda functions into MCP-compatible tools, Identity for access management, Memory for managed memory, built-in Code Interpreter, Browser and Web Search tools, OpenTelemetry-based observability, built-in and custom evaluators, and Cedar policies for fine-grained access control.

### Can you provide a tutorial for using Amazon AgentCore?

The repository is itself a tutorial collection. The README points to getting-started/ for a first agent using the AgentCore CLI, with separate python/ and typescript/ directories, and links a video walkthrough for building a first production-ready agent.

### What are AWS samples?

The repository does not define the term. Based on its own structure, this particular samples repository is a set of runnable examples and tutorials rather than a library, and its pyproject.toml explicitly describes it as a samples and tutorial repository with relaxed lint rules for tutorial-style code.

## Sources

- [awslabs/agentcore-samples on GitHub](https://github.com/awslabs/agentcore-samples)
- [Issues](https://github.com/awslabs/agentcore-samples/issues)
- [License: Apache-2.0](https://github.com/awslabs/agentcore-samples/blob/main/LICENSE)
- [Project website](https://aws.amazon.com/bedrock/agentcore/)
- [README](https://github.com/awslabs/agentcore-samples/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/awslabs-agentcore-samples