Agents, Orchestration & MCP
Agent loops, when to use agents vs. workflows, and the role of the Model Context Protocol.
Before you start
- 018 multiple-choice questions, one correct answer each.
- 02Suggested time 10 minutes. The timer is a guide, not a cutoff.
- 03Use keys 1–4 to answer, arrows to move.
- 04You get a full explanation for every question at the end.
Study guide
Every question in the Agents, Orchestration & MCP test, with the correct answer and a full explanation. Agent loops, when to use agents vs. workflows, and the role of the Model Context Protocol. Use it to review before or after taking the timed quiz above — the answers are revealed here, so take the quiz first if you want an honest score.
Show all 8questions, answers & explanations
- AG-01 · Question 1 of 8
What primarily distinguishes an 'agent' from a fixed 'workflow'?
- AAgents use larger models
- BAgents dynamically direct their own steps and tool use; workflows follow predefined paths Correct answer
- CWorkflows cannot call tools
- DAgents never use tools
Why: Workflows orchestrate LLMs and tools along predefined code paths. Agents let the model decide the steps and tool usage dynamically. Prefer the simplest option that works; reach for agents only when flexible, model-driven control is genuinely needed.
- AG-02 · Question 2 of 8
What problem does the Model Context Protocol (MCP) solve?
- AIt compresses prompts to save tokens
- BIt is a standard, open protocol for connecting models to external tools and data sources Correct answer
- CIt is a fine-tuning format
- DIt replaces the Messages API
Why: MCP is an open standard that gives models a consistent way to connect to external tools, data, and systems, so integrations are reusable across clients instead of bespoke per app.
- AG-03 · Question 3 of 8
Which guidance best reflects good agent design discipline?
- AAlways build a multi-agent system for any task
- BStart simple; add agentic complexity only when simpler solutions fall short Correct answer
- CAvoid tools in agents
- DMaximize the number of autonomous steps
Why: Favor the simplest design that meets requirements. Added autonomy increases cost, latency, and failure surface, so introduce agentic complexity only when a workflow or single call cannot do the job.
- AG-04 · Question 4 of 8
In an agent loop, what most reliably prevents runaway execution?
- ASetting `temperature` to 0
- BA maximum iteration / step budget and clear stopping conditions Correct answer
- CRemoving all tools
- DUsing a longer system prompt
Why: Bounded loops need explicit stop conditions and a step/iteration cap (plus cost and time limits) so a misbehaving agent cannot loop indefinitely. This is a core reliability and safety control.
- AG-05 · Question 5 of 8
Why give an agent a structured way to verify its own progress (e.g. tests, checks, ground-truth lookups)?
- AIt is required by the API
- BVerification signals let the agent course-correct instead of compounding errors Correct answer
- CIt reduces the context window
- DIt disables tool use
Why: Agents perform far better when they can check results against an objective signal (tests passing, a validator, a lookup). Feedback lets them recover from mistakes rather than confidently proceeding on a wrong path.
- AG-06 · Question 6 of 8
In the orchestrator-workers pattern, what is the orchestrator's main job?
- ATo execute every subtask itself
- BTo break a task into subtasks, delegate them to worker calls, and synthesize the results Correct answer
- CTo replace the need for any tools
- DTo fine-tune the worker models
Why: An orchestrator LLM decomposes a complex task, dispatches subtasks to worker calls (often in parallel), and combines their outputs. It is well suited to problems whose subtasks are not known in advance and must be planned dynamically.
- AG-07 · Question 7 of 8
An MCP server exposes capabilities to a client. Which is a typical primitive an MCP server provides?
- AModel weights for fine-tuning
- BTools (callable functions), resources (data), and prompts Correct answer
- CA replacement billing system
- DGPU scheduling
Why: MCP servers expose tools the model can call, resources it can read, and reusable prompts, all over a standard protocol. This lets one integration be reused across any MCP-compatible client instead of rebuilt per application.
- AG-08 · Question 8 of 8
For a task with well-understood, fixed steps where each step's output feeds the next, which design is most appropriate?
- AA fully autonomous agent
- BPrompt chaining: a fixed workflow that decomposes the task into sequential LLM calls Correct answer
- CA multi-agent swarm
- DA single call with maximum `max_tokens`
Why: When the steps are known and sequential, prompt chaining (a fixed workflow passing each call's output to the next, optionally with checks between) is simpler, cheaper, and more reliable than handing dynamic control to an autonomous agent.