ExamPark
Course menuMLA-C02

Exams / MLA-C02 / Study guide / Agents and agentic workflows

Agents and agentic workflows: MLA-C02 study notes and practice questions

What to know about Agents and agentic workflows for the MLA-C02 exam, from the Deployment and Orchestration of ML and AI Workflows domain (24% of the exam guide), then 1 free question to test it. Part of the MLA-C02 study guide.

Last verified against current AWS documentation ·How we write our questions

What to know for the exam

Agent questions describe a task that needs several steps or calls to company systems, and ask how to build or run it with managed services.

Amazon Bedrock Agents

  • Instructions tell the agent its role.
  • Action groups define tools with an OpenAPI schema or function details, run by Lambda (or returned to your app with return of control).
  • Knowledge bases give the agent retrieval.
  • Multi-agent collaboration lets a supervisor agent delegate to specialists.
  • Traces show each orchestration step, tool call and observation.
  • Aliases and versions support safe releases.
  • Session state and memory keep context across turns.

Amazon Bedrock AgentCore

A set of managed services to run agents built with any framework (such as Strands Agents, LangGraph or CrewAI) and any model:

Service Does
Runtime Serverless hosting for agents with session isolation and long-running tasks
Gateway Turns APIs (OpenAPI), Lambda functions and services into MCP-compatible tools behind one managed endpoint with authentication
Memory Short-term and long-term memory across sessions
Identity Manages inbound authentication and the agent’s access to third-party services
Code Interpreter / Browser Sandboxed tools for code and web tasks
Observability Traces and metrics in CloudWatch

Keeping agents safe

  • Least-privilege IAM roles for tools; the tool itself should check what it’s allowed to change.
  • Guardrails, including prompt attack filters, on inputs and outputs.
  • Human confirmation for risky actions.
  • Limit the number of steps and tools.

Common traps

  • Hosting your own MCP server fleet when AgentCore Gateway manages it.
  • Relying on the system prompt alone to stop harmful actions.
  • Adding more prompt text to debug failures instead of reading traces.

Try the 1 practice question

Practice Agents and agentic workflowsTake the free practice exam

Question 1

Medium ·Link to this question (question 1)

A public library is building a patron assistant agent with an open-source agent framework, hosted on Amazon Bedrock AgentCore Runtime. The agent must look up patron records and place holds through two existing REST services that have OpenAPI specifications, and renew loans through an existing AWS Lambda function.

The team wants the agent to discover and call all three as tools through one Model Context Protocol (MCP) endpoint that authenticates the agent. It does not want to write, host or patch its own MCP server.

Which solution meets these requirements?

Choose one answer.

  1. A. Build a custom MCP server in a container on Amazon ECS that wraps the two REST services and the Lambda function, and connect the agent to it.
  2. B. Create an Amazon Bedrock knowledge base over the two OpenAPI specifications, so that the agent can retrieve the API details and build the HTTP calls itself.
  3. C. Put an Amazon API Gateway REST API in front of the services and the Lambda function, and list the endpoint URLs in the agent's system prompt.
  4. D. Create an AgentCore Gateway with targets for the two OpenAPI specifications and the Lambda function, and point the agent at the gateway's MCP endpoint.
Show the answer and explanation

Answer: D

  • A. Wrong: A custom MCP server can expose the same tools, but the team would write, host and patch it, which the stem rules out. A managed gateway provides the MCP endpoint and authentication without that work.
  • B. Wrong: A knowledge base retrieves text for answers, and it does not call APIs or authenticate the agent. The agent would still have no managed tool endpoint for the lookups, holds and renewals.
  • C. Wrong: API Gateway exposes HTTP endpoints but not MCP tools, so the agent gets no tool definitions to discover. URLs in a prompt also leave the request format and authentication to the model.
  • D. Right: Correct. AgentCore Gateway converts APIs described by OpenAPI specifications and Lambda functions into MCP-compatible tools behind one endpoint, and it handles inbound authentication for the agent. The library writes and hosts no MCP server. Learn more: Amazon Bedrock AgentCore Gateway (opens in a new tab)

Explanation

AgentCore Gateway turns existing APIs, defined by OpenAPI specifications, and Lambda functions into MCP-compatible tools that an agent discovers and calls through one managed endpoint with inbound authentication. A self-hosted MCP server adds the operations the team wants to avoid, a knowledge base does not call APIs, and URLs in a prompt give the agent no tool definitions.

Study

Want more on Agents and agentic workflows? The 5 full-length practice sets have 11 more Agents and agentic workflows questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets

Study Agents and agentic workflows

More Deployment and Orchestration of ML and AI Workflows topics