AWS AgentCore Blueprints

Name: AWS AgentCore Blueprints
Author: Toc Consulting

Amazon Bedrock AgentCore, drawn as an engineer's book: every service explained in plain language, then done hands-on. 36 blueprint chapters in 5 parts, validated against the AWS documentation and the official AgentCore samples.

36 CHAPTERS 5 PARTS SOURCE-VERIFIED · REV 2026.07

Open the book: Chapter 01 CONTENTS

Chapters are 1400px engineering sheets, best read on a desktop screen. Use ← → keys to turn pages inside the book.

CH 01–36

PART I

Foundations

What AgentCore is, the words it uses, the mistakes to avoid, and what happens inside an agent.

01AgentCore Overview, Plain LanguageWhat Amazon Bedrock AgentCore is in plain words: the thirteen AWS services for building, hosting, and operating AI agents, and the two protocols (MCP and A2A) that connect them.CONCEPT 02AgentCore, Key Terms in Plain LanguagePlain-language definitions of the words behind AWS AgentCore: agent, tool, model, session, microVM, MCP, A2A, and more.CONCEPT 03AgentCore, Common Mistakes ExplainedThe most common mistakes teams make with Amazon Bedrock AgentCore, from ARM64 builds to idle timeouts and memory namespaces, and what to do instead.CONCEPT 04Inside an AgentCore Agent, BlueprintWhat actually happens inside an AgentCore agent: the agent loop, the harness, and how Strands Agents powers the managed harness.CONCEPT 05Harness, Practical 1, Strands PathBuild your first agent with Strands Agents step by step: install, define tools with the tool decorator, run the agent loop, and connect MCP tools.PRACTICAL 06Harness, Practical 2, Managed HarnessRun an agent with the AgentCore managed harness: the CLI path, the API path, and the boto3 path, with streaming events and stop reasons.PRACTICAL

PART II

Core Services

Runtime, Memory, Gateway, and Identity: hosting, remembering, tools, and secure access.

07AgentCore Runtime, Plain LanguageHow AgentCore Runtime hosts agents serverlessly: microVM session isolation, the session lifecycle, versioned endpoints, and the protocol contract.CONCEPT 08AgentCore Runtime, PracticalDeploy an agent to AgentCore Runtime hands-on with the AgentCore CLI: create, dev, deploy, invoke, and call it from code with boto3.PRACTICAL 09AgentCore Memory, Plain LanguageHow AgentCore Memory works: short-term conversation storage and long-term extracted facts, so agents remember across sessions.CONCEPT 10AgentCore Memory, PracticalUse AgentCore Memory hands-on: create a memory resource, add strategies, write conversation turns, and search what the agent learned.PRACTICAL 11AgentCore Gateway, Plain LanguageHow AgentCore Gateway turns APIs, Lambda functions, and MCP servers into tools agents can call, behind one secure MCP endpoint.CONCEPT 12AgentCore Gateway, PracticalSet up AgentCore Gateway hands-on: create a gateway, add a Lambda target, and call the tools from a Strands agent over MCP.PRACTICAL 13AgentCore Identity, Plain LanguageHow AgentCore Identity secures agents: inbound auth for who may call, outbound OAuth for what the agent may access, and the token vault.CONCEPT 14AgentCore Identity, PracticalConfigure AgentCore Identity hands-on: credential providers, OAuth 2LO and 3LO flows, and the requires_access_token decorator.PRACTICAL

PART III

Tools and Operations

Code Interpreter, Browser, Observability, and Evaluations: acting safely and watching quality.

15AgentCore Code Interpreter, Plain LanguageHow the AgentCore Code Interpreter gives agents a safe, isolated cloud sandbox to run Python, JavaScript, and TypeScript.CONCEPT 16AgentCore Code Interpreter, PracticalUse the Code Interpreter hands-on: start sessions, execute code with the SDK and boto3, and clean up.PRACTICAL 17AgentCore Browser, Plain LanguageHow the AgentCore Browser gives agents a managed cloud Chrome browser with live view, human takeover, and session replay.CONCEPT 18AgentCore Browser, PracticalDrive the AgentCore Browser hands-on with Strands: sessions, live view in the console, and the IAM permissions you need.PRACTICAL 19AgentCore Observability, Plain LanguageHow AgentCore Observability records every agent step as sessions, traces, and spans in CloudWatch, built on OpenTelemetry.CONCEPT 20AgentCore Observability, PracticalEnable observability hands-on: CloudWatch Transaction Search, auto-instrumented runtime agents, and ADOT for self-hosted agents.PRACTICAL 21AgentCore Evaluations, Plain LanguageHow AgentCore Evaluations scores agent quality automatically with built-in and custom LLM-as-a-Judge evaluators, on demand or continuously.CONCEPT 22AgentCore Evaluations, PracticalRun AgentCore Evaluations hands-on: add evaluators, grade past runs, set up online evaluation, and watch scores in CloudWatch.PRACTICAL

PART IV

Governance and Transactions

Policy, Registry, and Payments: rules, catalogs, and agents that spend real money.

23AgentCore Policy, Plain LanguageHow AgentCore Policy puts a deterministic Cedar-based checkpoint in front of every tool call an agent makes through Gateway.CONCEPT 24AgentCore Policy, PracticalWrite and attach AgentCore policies hands-on: generate Cedar from plain English, enforce at the gateway, and test what gets blocked.PRACTICAL 25AgentCore Registry, Plain LanguageHow AWS Agent Registry (preview) gives teams a shared, searchable catalog of approved agents, tools, and MCP servers.CONCEPT 26AgentCore Registry, PracticalUse the Agent Registry hands-on: create a registry, publish and approve records, and search them from code.PRACTICAL 27AgentCore Payments, Plain LanguageHow AgentCore Payments (preview) lets agents pay per use with stablecoin over the open x402 standard, inside spending limits you set.CONCEPT 28AgentCore Payments, PracticalSet up agent payments hands-on: credential providers, payment manager, wallet, budgeted sessions, and the Strands payments plugin.PRACTICAL

PART V

Multi-Agent and Architecture

A2A, architecture patterns, multi-tenancy, security, org-scale governance, and multi-region.

29A2A and Multi-Agent, Plain LanguageHow agents work together: the open A2A protocol for cross-team agents on AgentCore Runtime, and in-process multi-agent patterns in Strands.CONCEPT 30A2A and Multi-Agent, PracticalBuild an A2A agent hands-on: create with the CLI, serve on port 9000, read the agent card, and call it with the A2A client.PRACTICAL 31Architecture Patterns, Plain LanguageA decision guide for agent architecture on AgentCore: one agent or several, sequential vs hierarchical vs peer-to-peer, chat vs event triggers.CONCEPT 32Multi-Tenant Agents, Plain LanguageHow to serve many customers safely from one agent product on AgentCore: silo, pool, and bridge isolation models, and per-tenant cost control.CONCEPT 33Agent Security and Threat Model, Plain LanguageA practical threat model for AI agents on AWS: the OWASP Top 10 for LLM applications mapped to AgentCore controls and to what remains your job.CONCEPT 34Governance at Org Scale, Plain LanguageHow to govern many agent teams across AWS accounts: SCPs and Bedrock policies, a central registry, Cedar policies, and one observability pane.CONCEPT 35Governance at Org Scale, PracticalSet up cross-account agent observability hands-on: CloudWatch OAM sink and links via CloudFormation StackSets across your organization.PRACTICAL 36Multi-Region and Distribution, Plain LanguageHow to run agents across AWS Regions on AgentCore: the 20 available Regions, cross-region inference for residency, latency, and failover design.CONCEPT

Written by Tarek Cheikh, Toc Consulting. I researched, drew, and verified every chapter against the AWS AgentCore Developer Guide and the official amazon-bedrock-agentcore-samples repository (revision of July 2026). AgentCore evolves fast: always confirm limits and preview statuses against the current AWS documentation before building.