Topic 05 Practice Test 2 covers Agentic AI, Tool Calling, Human Review, Strands, Lambda, ECS, and MCP in Domain 2 (Task 2.1). For broader exam preparation, review the AIP-C01 Exam Dumps. Every option includes focused technical reasoning tied to the tested decision.
Question 1
A claims agent must resume a multi-hour case after the application process restarts without losing which external actions already completed. What should the team implement? Choose ONE.
- Strands Agents with explicit session state
- Lambda action implementation
- Application-managed durable state
- MCP client library
Correct Answer: C
Correct Answer
Answer C is correct because Application-managed durable state — for this design, stores durable workflow/session state outside, for this workload, the model so progress survives, for this workload, process restarts. Fits when a long-running agent task must, for the stated need, resume reliably after infrastructure interruption. Constraint: persist long-running workflow progress outside the model process. Application-managed durable state is the closest fit.
Incorrect Answers
Answer A is incorrect because Strands Agents with explicit session state — at this decision point, uses the Strands agent framework, within this design, while preserving the state needed, within this design, across turns or workflow steps. Typical use: an agent must retain task/session, in the described flow, context while coordinating tools. Goal: persist long-running workflow progress outside the model process. Strands Agents with explicit session state does not provide it.
Answer B is incorrect because Lambda action implementation — in the current architecture, runs bounded custom business logic, in the described flow, for an agent tool or, in the described flow, action group in a serverless, in the described flow, function. Applies where a tool action is lightweight,, in this case, stateless, and fits Lambda execution, in this case, constraints. Target outcome: persist long-running workflow progress outside the model process. Lambda action implementation misses that outcome.
Answer D is incorrect because MCP client library — within this implementation, gives agents a consistent client-side, for this decision, mechanism for discovering and invoking, for this decision, MCP-exposed tools. Strong fit when multiple MCP servers should be, within this design, accessed through a common client, within this design, pattern. Operational need: persist long-running workflow progress outside the model process. MCP client library is secondary here.
Question 2
A workflow calls a flaky third-party service that usually succeeds on a second or third attempt; retries must be delayed and bounded. Which option most directly addresses the requirement? Choose ONE.
- Circuit breaker around failing tool
- Lambda timeout guard
- Step Functions retry with backoff
- Compensating workflow
Correct Answer: C
Correct Answer
Answer C is correct because Step Functions retry with backoff — at this decision point, retries transient workflow steps according, within this design, to explicit retry policy and, within this design, delay rather than blindly repeating, within this design, immediately. Typical use: a recoverable dependency error should, in the described flow, be retried in a controlled, in the described flow, way. Goal: retry transient failures with controlled backoff. Step Functions retry with backoff provides that behavior.
Incorrect Answers
Answer A is incorrect because Circuit breaker around failing tool — under these constraints, temporarily stops repeated calls to, at this layer, a failing dependency after an, at this layer, error threshold and allows controlled, at this layer, recovery. Choose it when a repeatedly failing downstream tool, under these constraints, could otherwise amplify latency and, under these constraints, cost. Requirement: retry transient failures with controlled backoff. Circuit breaker around failing tool is not the direct fit.
Answer B is incorrect because Lambda timeout guard — in the current architecture, enforces bounded execution around a, in the described flow, tool or model-supporting function so, in the described flow, stalled work cannot run indefinitely. Applies where a lightweight serverless action needs, in this case, a clear execution timeout. Target outcome: retry transient failures with controlled backoff. Lambda timeout guard misses that outcome.
Answer D is incorrect because Compensating workflow — in this scenario, defines explicit reversal or remediation, in this case, steps when a multi-step agentic, in this case, transaction partially succeeds. Useful when a later tool failure must, at this layer, safely unwind earlier side effects. Need: retry transient failures with controlled backoff. Compensating workflow solves another problem.
Question 3
A payment tool may be retried after a network timeout, but a retry must not create a second charge. Which design best satisfies this requirement? Choose ONE.
- Circuit breaker around failing tool
- Idempotent tool operation
- Lambda parameter validation
- Step Functions retry with backoff
Correct Answer: B
Correct Answer
Answer B is correct because Idempotent tool operation — at this decision point, uses an idempotency token or, within this design, state check so a retried, within this design, action does not create duplicate, within this design, side effects. Typical use: workflow retries could otherwise repeat, in the described flow, a charge, ticket, or update. Goal: make a retried side-effecting operation safe from duplication. Idempotent tool operation provides that behavior.
Incorrect Answers
Answer A is incorrect because A circuit breaker can stop calls to a persistently failing dependency after an error threshold, reducing load while that dependency recovers. It does not deduplicate side effects that may already have succeeded before a timeout. The payment operation must be idempotent so a retry cannot create a second charge.
Answer C is incorrect because Lambda parameter validation — under these constraints, validates and normalizes tool inputs, at this layer, before business logic or downstream, at this layer, API execution. Choose it when tool calls need defensive checks, under these constraints, before side effects occur. Requirement: make a retried side-effecting operation safe from duplication. Lambda parameter validation is not the direct fit.
Answer D is incorrect because Step Functions retry with backoff controls when and how often a failed step is retried, but it does not make a payment operation safe to repeat. If the first request actually succeeded before the timeout, a later retry could still create another charge. An idempotent operation is required to prevent duplication.
Question 4
An HR agent should be able to read benefits data but must never discover or call payroll-adjustment tools. What is the most appropriate implementation choice? Choose ONE.
- Tool allowlist
- IAM least-privilege boundary
- Standardized function schema
- Per-tool IAM role
Correct Answer: A
Correct Answer
Answer A is correct because Tool allowlist — under these constraints, restricts the agent to an, at this layer, approved set of callable operations, at this layer, instead of exposing every available, at this layer, action. Choose it when the risk boundary is which, under these constraints, tools the model may invoke. Requirement: limit which tools the model may invoke at all. Tool allowlist addresses it precisely.
Incorrect Answers
Answer B is incorrect because IAM least-privilege boundary — at this decision point, limits the agent or tool, within this design, role to only the AWS, within this design, actions and resources required for, within this design, the task. Typical use: resource access must be constrained, in the described flow, independently of model behavior. Goal: limit which tools the model may invoke at all. IAM least-privilege boundary does not provide it.
Answer C is incorrect because A standardized function schema makes tool names and parameters unambiguous so the model can invoke approved functions correctly. It does not control whether payroll-adjustment tools are discoverable or callable by this HR agent. The requirement is an authorization or allowlist boundary that prevents those tools from being available at all.
Answer D is incorrect because Per-tool IAM role — in this scenario, assigns narrowly scoped permissions to, in this case, a specific tool integration rather, in this case, than sharing a broad execution, in this case, role. Useful when different tools have distinct resource-access requirements. Need: limit which tools the model may invoke at all. Per-tool IAM role solves another problem.
Question 5
A document tool needs S3 read access, while a ticketing tool needs only a specific DynamoDB table. The agent runtime should not share one broad role. Which approach should the developer use? Choose ONE.
- Cross-account IAM role
- Tool allowlist
- IAM least-privilege boundary
- Per-tool IAM role
Correct Answer: D
Correct Answer
Answer D is correct because Per-tool IAM role — for this design, assigns narrowly scoped permissions to, for this workload, a specific tool integration rather, for this workload, than sharing a broad execution, for this workload, role. Fits when different tools have distinct resource-access requirements. Constraint: separate AWS permissions by tool blast radius. Per-tool IAM role is the closest fit.
Incorrect Answers
Answer A is incorrect because Cross-account IAM role — in this scenario, provides temporary, scoped cross-account access, in this case, without sharing long-lived user credentials. Useful when a workload in one AWS, at this layer, account must access approved resources, at this layer, in another. Need: separate AWS permissions by tool blast radius. Cross-account IAM role solves another problem.
Answer B is incorrect because Tool allowlist — for the described workload, restricts the agent to an, for the stated need, approved set of callable operations, for the stated need, instead of exposing every available, for the stated need, action. Best suited when the risk boundary is which, for this decision, tools the model may invoke. Deciding factor: separate AWS permissions by tool blast radius. Tool allowlist addresses something else.
Answer C is incorrect because IAM least-privilege boundary — under these constraints, limits the agent or tool, at this layer, role to only the AWS, at this layer, actions and resources required for, at this layer, the task. Choose it when resource access must be constrained, under these constraints, independently of model behavior. Requirement: separate AWS permissions by tool blast radius. IAM least-privilege boundary is not the direct fit.
Question 6
An agent books travel in two systems; the first booking succeeds and the second fails. The design must safely cancel the first booking. What should the team implement? Choose ONE.
- Compensating workflow
- Idempotent tool operation
- Step Functions retry with backoff
- Circuit breaker around failing tool
Correct Answer: A
Correct Answer
Answer A is correct because Compensating workflow — for this design, defines explicit reversal or remediation, for this workload, steps when a multi-step agentic, for this workload, transaction partially succeeds. Fits when a later tool failure must, for the stated need, safely unwind earlier side effects. Constraint: undo earlier side effects after a later step fails. Compensating workflow is the closest fit.
Incorrect Answers
Answer B is incorrect because Idempotent tool operation — under these constraints, uses an idempotency token or, at this layer, state check so a retried, at this layer, action does not create duplicate, at this layer, side effects. Choose it when workflow retries could otherwise repeat, under these constraints, a charge, ticket, or update. Requirement: undo earlier side effects after a later step fails. Idempotent tool operation is not the direct fit.
Answer C is incorrect because Retry with backoff is appropriate when the failed second booking is transient and another attempt may succeed. It still does not address the required cleanup when the overall workflow ultimately fails. A compensating workflow must reverse the first successful booking so the two-system transaction does not remain partially completed.
Answer D is incorrect because A circuit breaker protects against a dependency that is repeatedly failing by suppressing further calls for a period. It does not reverse a successful booking in another system after a later step fails. This workflow needs compensating logic that explicitly cancels the first booking when the second booking cannot be completed.
Question 7
An agent iteratively plans, acts, observes results, and decides the next step. The team wants the loop represented as explicit workflow states. Which option most directly addresses the requirement? Choose ONE.
- Strands Agents with explicit session state
- Model-selection framework
- MCP client library
- Step Functions orchestration
Correct Answer: D
Correct Answer
Answer D is correct because Step Functions orchestration — at this decision point, models multi-step workflow control, branching,, within this design, retries, approvals, and stopping conditions, within this design, explicitly. Typical use: an AI workflow needs deterministic, in the described flow, orchestration around model or tool, in the described flow, calls. Goal: orchestrate structured reasoning and action states deterministically. Step Functions orchestration provides that behavior.
Incorrect Answers
Answer A is incorrect because Strands Agents with explicit session state — under these constraints, uses the Strands agent framework, at this layer, while preserving the state needed, at this layer, across turns or workflow steps. Choose it when an agent must retain task/session, under these constraints, context while coordinating tools. Requirement: orchestrate structured reasoning and action states deterministically. Strands Agents with explicit session state is not the direct fit.
Answer B is incorrect because Model-selection framework — under these constraints, applies policy or scoring logic, at this layer, to choose among candidate models, at this layer, at runtime. Choose it when the application must choose a, under these constraints, model based on measurable request, under these constraints, characteristics. Requirement: orchestrate structured reasoning and action states deterministically. Model-selection framework is not the direct fit.
Answer C is incorrect because MCP client library — in this scenario, gives agents a consistent client-side, in this case, mechanism for discovering and invoking, in this case, MCP-exposed tools. Appropriate if multiple MCP servers should be, at this layer, accessed through a common client, at this layer, pattern. The stem requires orchestrate structured reasoning and action states deterministically. MCP client library serves another purpose.
Question 8
A tool must be denied access to every S3 bucket except one approved prefix, even if a prompt attempts to request a broader path. Which design best satisfies this requirement? Choose ONE.
- IAM least-privilege boundary
- Lambda parameter validation
- Standardized function schema
- Tool allowlist
Correct Answer: A
Correct Answer
Answer A is correct because IAM least-privilege boundary — for the described workload, limits the agent or tool, for the stated need, role to only the AWS, for the stated need, actions and resources required for, for the stated need, the task. Best suited when resource access must be constrained, for this decision, independently of model behavior. Deciding factor: enforce resource scope against prompt-driven privilege expansion. IAM least-privilege boundary fits this case.
Incorrect Answers
Answer B is incorrect because Lambda parameter validation — for the described workload, validates and normalizes tool inputs, for the stated need, before business logic or downstream, for the stated need, API execution. Best suited when tool calls need defensive checks, for this decision, before side effects occur. Deciding factor: enforce resource scope against prompt-driven privilege expansion. Lambda parameter validation addresses something else.
Answer C is incorrect because Standardized function schema — for this operational need, defines tool names, parameters, types,, under these constraints, and required fields consistently so, under these constraints, the model can call tools, under these constraints, reliably. Architects use it when tool invocation errors are caused, for this workload, by ambiguous or inconsistent function, for this workload, definitions. Key need: enforce resource scope against prompt-driven privilege expansion. Standardized function schema does not satisfy it.
Answer D is incorrect because Tool allowlist — in the current architecture, restricts the agent to an, in the described flow, approved set of callable operations, in the described flow, instead of exposing every available, in the described flow, action. Applies where the risk boundary is which, in this case, tools the model may invoke. Target outcome: enforce resource scope against prompt-driven privilege expansion. Tool allowlist misses that outcome.
Question 9
The team wants to compare outputs from two models in parallel and select a response only if both satisfy independent policy checks. What is the most appropriate implementation choice? Choose ONE.
- AWS Agent Squad multi-agent coordination
- Custom result aggregation
- Specialist-model routing
- Model-selection framework
Correct Answer: B
Correct Answer
Answer B is correct because Custom result aggregation — in the current architecture, combines outputs from multiple model, in the described flow, or agent workers using explicit, in the described flow, application logic. Applies where several parallel specialist results must, in this case, be reconciled into one answer. Target outcome: reconcile parallel model outputs with explicit acceptance logic. Custom result aggregation supplies it directly.
Incorrect Answers
Answer A is incorrect because AWS Agent Squad multi-agent coordination — in the current architecture, coordinates multiple specialized agents so, in the described flow, each can contribute a bounded, in the described flow, capability to a larger task. Applies where a workflow benefits from several, in this case, specialist agents rather than one, in this case, monolithic agent. Target outcome: reconcile parallel model outputs with explicit acceptance logic. AWS Agent Squad multi-agent coordination misses that outcome.
Answer C is incorrect because Specialist-model routing — in the current architecture, routes a request to a, in the described flow, model selected for the task’s, in the described flow, capability, latency, or quality needs. Applies where different task classes benefit from, in this case, different model strengths. Target outcome: reconcile parallel model outputs with explicit acceptance logic. Specialist-model routing misses that outcome.
Answer D is incorrect because Model-selection framework — for the described workload, applies policy or scoring logic, for the stated need, to choose among candidate models, for the stated need, at runtime. Best suited when the application must choose a, for this decision, model based on measurable request, for this decision, characteristics. Deciding factor: reconcile parallel model outputs with explicit acceptance logic. Model-selection framework addresses something else.
Question 10
A dispatcher should select a coding-specialized model for code repair and a vision-capable model when an image is attached. Which approach should the developer use? Choose ONE.
- Specialist-model routing
- MCP for agent-tool interoperability
- Model-selection framework
- Custom result aggregation
Correct Answer: A
Correct Answer
Answer A is correct because Specialist-model routing — for this operational need, routes a request to a, under these constraints, model selected for the task’s, under these constraints, capability, latency, or quality needs. Architects use it when different task classes benefit from, for this workload, different model strengths. Key need: route based on modality and task specialization. Specialist-model routing aligns with it.
Incorrect Answers
Answer B is incorrect because MCP for agent-tool interoperability — under these constraints, uses the Model Context Protocol, at this layer, to expose tools through a, at this layer, consistent agent-to-tool interface. Choose it when tools should be reachable through, under these constraints, a standardized protocol instead of, under these constraints, bespoke integrations. Requirement: route based on modality and task specialization. MCP for agent-tool interoperability is not the direct fit.
Answer C is incorrect because Model-selection framework — in the current architecture, applies policy or scoring logic, in the described flow, to choose among candidate models, in the described flow, at runtime. Applies where the application must choose a, in this case, model based on measurable request, in this case, characteristics. Target outcome: route based on modality and task specialization. Model-selection framework misses that outcome.
Answer D is incorrect because Custom result aggregation — for this operational need, combines outputs from multiple model, under these constraints, or agent workers using explicit, under these constraints, application logic. Architects use it when several parallel specialist results must, for this workload, be reconciled into one answer. Key need: route based on modality and task specialization. Custom result aggregation does not satisfy it.
Question 11
A Bedrock agent has selected the correct order-status function, but customers sometimes omit the required order ID. The team wants the agent to ask for the missing value instead of inventing a parameter or invoking the function with incomplete input. What should the developer enable? Choose ONE.
- Add Lambda parameter validation
- Enable user confirmation for the action
- Use RETURN_CONTROL for the action group
- Enable user input for the action group
Correct Answer: D
Correct Answer
Answer D is correct because Enabling user input lets a Bedrock agent ask the end user for missing action parameters instead of fabricating a value or invoking the function incompletely. The order ID is required before the tool can run, so eliciting that value inside the agent flow directly solves the scenario.
Incorrect Answers
Answer A is incorrect because Lambda parameter validation can reject malformed or absent values before business logic runs, which protects the downstream operation. It cannot make the agent converse with the user to obtain the missing order ID, so it would detect the problem without providing the requested elicitation behavior.
Answer B is incorrect because User confirmation asks the user to confirm or deny an action before the function is invoked. That protects consent-sensitive operations, but it is not the mechanism for eliciting a missing required parameter. The order ID still has to be collected before the action can be called.
Answer C is incorrect because RETURN_CONTROL sends the selected action and elicited parameters back to the caller for application-managed execution. It is useful when the caller must perform the action itself, but it does not specifically make the agent request a missing required order ID from the user.
Question 12
A mobile app must capture a reviewer’s star rating and comment for each generated answer so feedback can feed later evaluation. Which option most directly addresses the requirement? Choose ONE.
- MCP client library
- Human approval in Step Functions
- API Gateway feedback endpoint
- Standardized function schema
Correct Answer: C
Correct Answer
Answer C is correct because API Gateway feedback endpoint — in the current architecture, provides an HTTP interface through, in the described flow, which applications can submit structured, in the described flow, human feedback or approval data. Applies where the workflow needs an external, in this case, feedback collection endpoint. Target outcome: collect human feedback through an application-facing API. API Gateway feedback endpoint supplies it directly.
Incorrect Answers
Answer A is incorrect because MCP client library — for this design, gives agents a consistent client-side, for this workload, mechanism for discovering and invoking, for this workload, MCP-exposed tools. Supports designs where multiple MCP servers should be, for the stated need, accessed through a common client, for the stated need, pattern. The workload needs collect human feedback through an application-facing API. MCP client library maps elsewhere.
Answer B is incorrect because Human approval in Step Functions — under these constraints, pauses an orchestrated workflow for, at this layer, an explicit human review or, at this layer, approval before continuing. Choose it when high-impact actions require human judgment, under these constraints, before execution. Requirement: collect human feedback through an application-facing API. Human approval in Step Functions is not the direct fit.
Answer D is incorrect because Standardized function schema — within this implementation, defines tool names, parameters, types,, for this decision, and required fields consistently so, for this decision, the model can call tools, for this decision, reliably. Strong fit when tool invocation errors are caused, within this design, by ambiguous or inconsistent function, within this design, definitions. Operational need: collect human feedback through an application-facing API. Standardized function schema is secondary here.
Question 13
A Bedrock action group should hand an order request back to the application because fulfillment occurs in a private ERP integration managed by the application. Which design best satisfies this requirement? Choose ONE.
- API Gateway feedback endpoint
- RETURN_CONTROL action group
- Lambda action implementation
- MCP server on Amazon ECS
Correct Answer: B
Correct Answer
Answer B is correct because RETURN_CONTROL causes the Bedrock agent to return the chosen action and elicited parameters to the caller instead of invoking a Lambda fulfillment function. That lets the application use its private ERP integration and existing connectivity to execute the order request, which directly matches the requirement for application-managed fulfillment.
Incorrect Answers
Answer A is incorrect because API Gateway feedback endpoint — for this operational need, provides an HTTP interface through, under these constraints, which applications can submit structured, under these constraints, human feedback or approval data. Architects use it when the workflow needs an external, for this workload, feedback collection endpoint. Key need: let the caller execute the business action outside the action group. API Gateway feedback endpoint does not satisfy it.
Answer C is incorrect because A Lambda action implementation executes the fulfillment logic from an AWS Lambda function associated with the action group. The stated design instead requires the calling application to perform fulfillment through its private ERP integration. RETURN_CONTROL is the direct mechanism for handing the action and parameters back to that caller.
Answer D is incorrect because MCP server on Amazon ECS — under these constraints, hosts an MCP server in, at this layer, containers suitable for longer-lived, more, at this layer, complex, or dependency-heavy tool services. Choose it when the tool server needs persistent, under these constraints, process behavior or containerized dependencies. Requirement: let the caller execute the business action outside the action group. MCP server on Amazon ECS is not the direct fit.
Question 14
A custom tool is invoked with dates, region codes, and account IDs. The team wants malformed values rejected with clear errors before downstream APIs run. What is the most appropriate implementation choice? Choose ONE.
- Lambda timeout guard
- Lambda parameter validation
- Idempotent tool operation
- Standardized function schema
Correct Answer: B
Correct Answer
Answer B is correct because Lambda parameter validation — in the current architecture, validates and normalizes tool inputs, in the described flow, before business logic or downstream, in the described flow, API execution. Applies where tool calls need defensive checks, in this case, before side effects occur. Target outcome: defensively validate tool arguments. Lambda parameter validation supplies it directly.
Incorrect Answers
Answer A is incorrect because Lambda timeout guard — for this operational need, enforces bounded execution around a, under these constraints, tool or model-supporting function so, under these constraints, stalled work cannot run indefinitely. Architects use it when a lightweight serverless action needs, for this workload, a clear execution timeout. Key need: defensively validate tool arguments. Lambda timeout guard does not satisfy it.
Answer C is incorrect because Idempotent tool operation — for the described workload, uses an idempotency token or, for the stated need, state check so a retried, for the stated need, action does not create duplicate, for the stated need, side effects. Best suited when workflow retries could otherwise repeat, for this decision, a charge, ticket, or update. Deciding factor: defensively validate tool arguments. Idempotent tool operation addresses something else.
Answer D is incorrect because Standardized function schema — in this scenario, defines tool names, parameters, types,, in this case, and required fields consistently so, in this case, the model can call tools, in this case, reliably. Appropriate if tool invocation errors are caused, at this layer, by ambiguous or inconsistent function, at this layer, definitions. The stem requires defensively validate tool arguments. Standardized function schema serves another purpose.
Question 15
A shared tool registry exposes inconsistent parameter names. The agent often calls the right capability with the wrong argument shape. Which approach should the developer use? Choose ONE.
- Lambda parameter validation
- Tool allowlist
- MCP client library
- Standardized function schema
Correct Answer: D
Correct Answer
Answer D is correct because Standardized function schema — for this design, defines tool names, parameters, types,, for this workload, and required fields consistently so, for this workload, the model can call tools, for this workload, reliably. Supports designs where tool invocation errors are caused, for the stated need, by ambiguous or inconsistent function, for the stated need, definitions. The workload needs normalize tool contracts so function calling is reliable. Standardized function schema maps to it.
Incorrect Answers
Answer A is incorrect because Lambda parameter validation — for this operational need, validates and normalizes tool inputs, under these constraints, before business logic or downstream, under these constraints, API execution. Architects use it when tool calls need defensive checks, for this workload, before side effects occur. Key need: normalize tool contracts so function calling is reliable. Lambda parameter validation does not satisfy it.
Answer B is incorrect because Tool allowlist — within this implementation, restricts the agent to an, for this decision, approved set of callable operations, for this decision, instead of exposing every available, for this decision, action. Strong fit when the risk boundary is which, within this design, tools the model may invoke. Operational need: normalize tool contracts so function calling is reliable. Tool allowlist is secondary here.
Answer C is incorrect because MCP client library — at this decision point, gives agents a consistent client-side, within this design, mechanism for discovering and invoking, within this design, MCP-exposed tools. Use it where multiple MCP servers should be, in the described flow, accessed through a common client, in the described flow, pattern. The design requires normalize tool contracts so function calling is reliable. MCP client library is not the direct choice.
Question 16
A stateless MCP endpoint only looks up a small reference table and is used a few times per hour. What should the team implement? Choose ONE.
- Lambda action implementation
- Application-managed durable state
- Stateless MCP server on Lambda
- MCP server on Amazon ECS
Correct Answer: C
Correct Answer
Answer C is correct because Stateless MCP server on Lambda — at this decision point, hosts lightweight, stateless MCP tool, within this design, access in Lambda with minimal, within this design, server management. Typical use: the MCP workload is short-lived,, in the described flow, stateless, and event-driven. Goal: use serverless hosting for lightweight intermittent MCP traffic. Stateless MCP server on Lambda provides that behavior.
Incorrect Answers
Answer A is incorrect because Lambda action implementation — within this implementation, runs bounded custom business logic, for this decision, for an agent tool or, for this decision, action group in a serverless, for this decision, function. Strong fit when a tool action is lightweight,, within this design, stateless, and fits Lambda execution, within this design, constraints. Operational need: use serverless hosting for lightweight intermittent MCP traffic. Lambda action implementation is secondary here.
Answer B is incorrect because Application-managed durable state — at this decision point, stores durable workflow/session state outside, within this design, the model so progress survives, within this design, process restarts. Typical use: a long-running agent task must, in the described flow, resume reliably after infrastructure interruption. Goal: use serverless hosting for lightweight intermittent MCP traffic. Application-managed durable state does not provide it.
Answer D is incorrect because MCP server on Amazon ECS — for the described workload, hosts an MCP server in, for the stated need, containers suitable for longer-lived, more, for the stated need, complex, or dependency-heavy tool services. Best suited when the tool server needs persistent, for this decision, process behavior or containerized dependencies. Deciding factor: use serverless hosting for lightweight intermittent MCP traffic. MCP server on Amazon ECS addresses something else.
Question 17
An MCP service runs a native PDF parser and maintains a warm pool of expensive dependencies to avoid repeated startup cost. Which option most directly addresses the requirement? Choose ONE.
- MCP client library
- Stateless MCP server on Lambda
- AWS Agent Squad multi-agent coordination
- MCP server on Amazon ECS
Correct Answer: D
Correct Answer
Answer D is correct because MCP server on Amazon ECS — in the current architecture, hosts an MCP server in, in the described flow, containers suitable for longer-lived, more, in the described flow, complex, or dependency-heavy tool services. Applies where the tool server needs persistent, in this case, process behavior or containerized dependencies. Target outcome: use container hosting for a complex persistent MCP service. MCP server on Amazon ECS supplies it directly.
Incorrect Answers
Answer A is incorrect because MCP client library — under these constraints, gives agents a consistent client-side, at this layer, mechanism for discovering and invoking, at this layer, MCP-exposed tools. Makes sense when multiple MCP servers should be, under these constraints, accessed through a common client, under these constraints, pattern. The case depends on use container hosting for a complex persistent MCP service. MCP client library handles another concern.
Answer B is incorrect because Stateless MCP server on Lambda — under these constraints, hosts lightweight, stateless MCP tool, at this layer, access in Lambda with minimal, at this layer, server management. Choose it when the MCP workload is short-lived,, under these constraints, stateless, and event-driven. Requirement: use container hosting for a complex persistent MCP service. Stateless MCP server on Lambda is not the direct fit.
Answer C is incorrect because Multi-agent coordination divides a larger task among specialized agents and combines their work. It does not provide the runtime characteristics needed by a native PDF parser with expensive warm dependencies. A long-lived container service on Amazon ECS is better suited to persistent processes, native libraries, and controlled resource allocation.
Question 18
An internal framework should let the same agent connect to finance, HR, and support MCP servers without implementing three protocol stacks. Which design best satisfies this requirement? Choose ONE.
- Standardized function schema
- MCP client library
- Specialist-model routing
- MCP for agent-tool interoperability
Correct Answer: B
Correct Answer
Answer B is correct because MCP client library — for the described workload, gives agents a consistent client-side, for the stated need, mechanism for discovering and invoking, for the stated need, MCP-exposed tools. Intended when multiple MCP servers should be, for this decision, accessed through a common client, for this decision, pattern. The architecture needs access multiple MCP servers through a consistent client library. MCP client library fits best.
Incorrect Answers
Answer A is incorrect because A standardized function schema describes the contract for an individual callable tool, including its parameters and types. It does not remove the need to implement separate protocol clients for multiple servers. An MCP client library provides the common discovery, connection, and invocation behavior required across the finance, HR, and support MCP servers.
Answer C is incorrect because Specialist-model routing — within this implementation, routes a request to a, for this decision, model selected for the task’s, for this decision, capability, latency, or quality needs. Strong fit when different task classes benefit from, within this design, different model strengths. Operational need: access multiple MCP servers through a consistent client library. Specialist-model routing is secondary here.
Answer D is incorrect because MCP for agent-tool interoperability — for the described workload, uses the Model Context Protocol, for the stated need, to expose tools through a, for the stated need, consistent agent-to-tool interface. Best suited when tools should be reachable through, for this decision, a standardized protocol instead of, for this decision, bespoke integrations. Deciding factor: access multiple MCP servers through a consistent client library. MCP for agent-tool interoperability addresses something else.
Question 19
A remediation agent has a maximum of five autonomous action cycles. After that, it must stop and request human assistance. What is the most appropriate implementation choice? Choose ONE.
- Human approval in Step Functions
- Circuit breaker around failing tool
- Explicit stopping condition
- Lambda timeout guard
Correct Answer: C
Correct Answer
Answer C is correct because Explicit stopping condition — at this decision point, terminates iterative reasoning or tool-use, within this design, loops when a defined completion,, within this design, risk, or iteration condition is, within this design, reached. Typical use: an autonomous loop must not, in the described flow, continue indefinitely. Goal: bound the number of autonomous reasoning/action iterations. Explicit stopping condition provides that behavior.
Incorrect Answers
Answer A is incorrect because Human approval in Step Functions — for the described workload, pauses an orchestrated workflow for, for the stated need, an explicit human review or, for the stated need, approval before continuing. Best suited when high-impact actions require human judgment, for this decision, before execution. Deciding factor: bound the number of autonomous reasoning/action iterations. Human approval in Step Functions addresses something else.
Answer B is incorrect because A circuit breaker limits calls to a failing external dependency after repeated errors. The remediation agent may be healthy and still exceed its allowed number of autonomous reasoning-and-action cycles. An explicit iteration or stopping condition should end the loop after five cycles and route the case for human assistance.
Answer D is incorrect because Lambda timeout guard — within this implementation, enforces bounded execution around a, for this decision, tool or model-supporting function so, for this decision, stalled work cannot run indefinitely. Strong fit when a lightweight serverless action needs, within this design, a clear execution timeout. Operational need: bound the number of autonomous reasoning/action iterations. Lambda timeout guard is secondary here.
Question 20
A downstream search service returns HTTP 503 intermittently. The workflow should wait, retry a few times, and then fail cleanly. Which approach should the developer use? Choose ONE.
- Step Functions retry with backoff
- Idempotent tool operation
- Circuit breaker around failing tool
- Lambda timeout guard
Correct Answer: A
Correct Answer
Answer A is correct because Step Functions retry with backoff is designed for transient failures such as intermittent HTTP 503 responses. The workflow can wait between attempts, limit the maximum retry count, and transition to a controlled failure path when attempts are exhausted. That directly satisfies the requirement to retry briefly and then fail cleanly.
Incorrect Answers
Answer B is incorrect because Idempotent tool operation — in the current architecture, uses an idempotency token or, in the described flow, state check so a retried, in the described flow, action does not create duplicate, in the described flow, side effects. Applies where workflow retries could otherwise repeat, in this case, a charge, ticket, or update. Target outcome: handle transient service failure without immediate repeated calls. Idempotent tool operation misses that outcome.
Answer C is incorrect because A circuit breaker is most useful when a dependency is persistently unhealthy and the application should suppress new calls across requests until recovery. This scenario explicitly calls for a small number of delayed retries for intermittent 503 responses and then a clean failure, making a bounded retry-with-backoff policy the more direct control.
Answer D is incorrect because Lambda timeout guard — in this scenario, enforces bounded execution around a, in this case, tool or model-supporting function so, in this case, stalled work cannot run indefinitely. Appropriate if a lightweight serverless action needs, at this layer, a clear execution timeout. The stem requires handle transient service failure without immediate repeated calls. Lambda timeout guard serves another purpose.
