Microsoft AB-620 Practice Questions with Explanations
Free Microsoft AB-620 practice questions. 20 of them, each with the correct answer, a full explanation, and the reason every other option is wrong.
AB-620 practice questions
Question 1
An agent calls a flow. The agent requires structured output values to be returned.
The agent receives unexpected or empty values.
You need to configure the agent so that data is exchanged correctly between the agent and the flow. What should you do?
- A. Validate parameter definitions.
- B. Change the trigger schema.
- C. Review the flow run history.
- D. Add a Parse JSON action inside the flow.
- E. Republish the agent.
Show answer and explanation ▾
Correct answer: A
When an agent receives empty or unexpected values from a flow, the usual cause is a mismatch between what the agent expects and what the flow declares. The flow's input parameters on its trigger and output parameters in its response to the agent define the contract, including names and data types. Validating those definitions, and confirming that the flow actually assigns values to each output and that the agent maps them to variables, corrects how data moves in both directions. Fixing the contract addresses the root cause instead of only observing the symptom.
Why the other options are wrong:
- B. Altering the trigger schema only affects inputs and can break the existing contract, without fixing the outputs that return empty values.
- C. Run history helps diagnose what happened in a run, but reviewing it does not configure the data exchange correctly.
- D. Parse JSON structures data inside the flow, yet values still reach the agent only through correctly defined output parameters.
- E. Republishing deploys the current configuration unchanged, so a parameter mismatch would persist.
Question 2
A company is building an agent in Copilot Studio.
The agent must meet the following requirements:
Responses must be grounded in approved internal content.
The agent must be able to retrieve information from configured data sources in a topic.
You need to configure the agent so that topic responses are grounded from custom data sources.
What should you configure?
- A. Enable analytics to review which topics users trigger most often.
- B. Add a custom prompt that tells the agent to answer only from company data.
- C. Configure knowledge sources in a generative answers node.
- D. Add SharePoint as a knowledge source only at the agent level.
- E. Increase the response length limit to improve answer completeness.
Show answer and explanation ▾
Correct answer: C
The requirement is for responses inside a topic to be grounded in approved internal content from configured data sources. A generative answers node placed in the topic searches the knowledge sources specified in its own settings and composes a response based only on what it finds. Configuring the node's knowledge sources lets the builder point that topic at the specific approved content, such as particular SharePoint sites, files, or other supported sources, so the answers stay grounded where the topic needs them.
Why the other options are wrong:
- A. Analytics report which topics users trigger, but they do not ground any response in company data.
- B. A prompt instruction alone cannot ground answers, because the model still needs knowledge sources to retrieve the approved content from.
- D. Agent-level knowledge serves general orchestration, while the requirement is to retrieve from configured sources within a topic, which the node's own knowledge settings provide.
- E. Response length affects how long an answer is, not whether it is grounded in approved content.
Question 3
A company is deploying an agent for employees.
The company has the following requirements:
Access must be restricted to internal employees only.
End-user authentication must be enforced.
You need to configure the agent based on the company requirements.
What should you do?
- A. Select Authenticate with Microsoft Entra ID for the agent.
- B. Enable Require Secure Access in the Web channel security.
- C. Publish the agent as a public website experience.
- D. Select Authenticate with personal account for the agent.
Show answer and explanation ▾
Correct answer: A
Restricting an agent to internal employees while enforcing end-user sign-in calls for authentication through Microsoft Entra ID. With this setting, users must sign in with their organizational work account, so only people in the company's tenant can use the agent, and their identity is available to the agent for secure access to data. It also supports internal channels such as Microsoft Teams and Microsoft 365 Copilot without extra identity configuration. This single setting satisfies both the access restriction and the authentication requirement.
Why the other options are wrong:
- B. Secure access for the web channel protects how the embed is loaded, but it does not require employees to sign in with organizational identities.
- C. A public website experience opens the agent to anyone, which contradicts the internal-only access requirement.
- D. Personal accounts are consumer identities, so they would not limit access to the organization's employees.
Question 4
An agent in Copilot Studio already exists in a development environment.
You need to prepare the agent so it can be moved to a production environment using the supported solution-based approach.
You need to create an exportable package that contains the existing agent.
Which two actions should you perform? Each correct answer presents part of the solution.
NOTE: Each correct selection is worth one point.
- A. Remove agent topics directly from the solution to prevent export errors.
- B. Convert the solution to a managed solution so it can be exported.
- C. Export the unmanaged solution that contains the agent.
- D. Publish the agent to a channel and export the channel configuration.
- E. Create a custom solution and add the existing agent to the solution.
Show answer and explanation ▾
Correct answer: C, E
Copilot Studio moves agents between environments by packaging them in solutions. The documented process starts by creating a custom solution in the solution explorer and using Add existing to place the existing agent in it, along with any required objects. You then open the list of unmanaged solutions, select the solution that contains the agent, and choose Export solution to produce the package that is imported into production. Solutions are unmanaged by default, and keeping the solution unmanaged is what allows the export, because a solution that has been changed to managed cannot be exported from that environment.
Why the other options are wrong:
- A. Removing agent components such as topics directly from the solution is exactly what causes export and import to fail, so it does not prevent errors.
- B. A managed solution cannot be exported, so changing the solution to managed blocks the export rather than enabling it; a new solution would be required.
- D. Publishing to a channel makes the agent available to users, and channel details do not travel with an exported agent, so this creates no transportable package.
Question 5
Note: This question is part of a series of questions that present the same scenario. Each question in the series contains a unique solution that might meet the stated goals. Some question sets might have more than one correct solution, while others might not have a correct solution.
After you answer a question in this section, you will NOT be able to return to it. As a result, these questions will not appear on the review screen.
An agent uses a flow that calls an external service which can occasionally fail or time out.
When a failure occurs, the agent must meet the following requirements:
Must not terminate silently.
Must send a notification containing the error details.
You need to configure the agent flow so that failures are handled in a controlled and predictable way.
Solution: Retry the failing action without handling failure.
Does the solution meet the goal?
- A. Yes
- B. No
Show answer and explanation ▾
Correct answer: B
Retrying the failing action can help with transient timeouts, but by itself it does nothing once the retries are exhausted. If the external service is still unavailable, the action fails and the run stops without any branch that reacts to the failure, so nothing tells anyone what went wrong. The goal requires that failures never end silently and that a notification with the error details is sent. Meeting that goal needs a failure path, such as an action configured to run after the failing step has failed or timed out, that sends the notification. A retry policy alone provides neither of these.
Why the other options are wrong:
- A. A retry policy only repeats the call; when every attempt fails, the run still ends without a failure branch or an error notification, so the requirements are not met.
Question 6
Note: This question is part of a series of questions that present the same scenario. Each question in the series contains a unique solution that might meet the stated goals. Some question sets might have more than one correct solution, while others might not have a correct solution.
After you answer a question in this section, you will NOT be able to return to it. As a result, these questions will not appear on the review screen.
An agent uses a flow that calls an external service which can occasionally fail or time out.
When a failure occurs, the agent must meet the following requirements:
Must not terminate silently.
Must send a notification containing the error details.
You need to configure the agent flow so that failures are handled in a controlled and predictable way.
Solution: Configure run after conditions for failure paths.
Does the solution meet the goal?
- A. Yes
- B. No
Show answer and explanation ▾
Correct answer: A
Run after settings let an action run only when a previous action has failed, timed out, been skipped or succeeded. By configuring a failure path that runs after the call to the external service has failed or timed out, the flow keeps control when the service misbehaves instead of simply stopping. That path can capture the error output of the failed step and send a notification containing those details, and it can then end the run with a defined status. This gives predictable, controlled handling of failures, so the agent never terminates silently and the required error notification is always sent.
Why the other options are wrong:
- B. Run after conditions are the mechanism that creates a dedicated failure branch, which can send the error notification, so rejecting this solution ignores how flows handle failed or timed-out actions.
Question 7
Note: This question is part of a series of questions that present the same scenario. Each question in the series contains a unique solution that might meet the stated goals. Some question sets might have more than one correct solution, while others might not have a correct solution.
After you answer a question in this section, you will NOT be able to return to it. As a result, these questions will not appear on the review screen.
An agent uses a flow that calls an external service which can occasionally fail or time out.
When a failure occurs, the agent must meet the following requirements:
Must not terminate silently.
Must send a notification containing the error details.
You need to configure the agent flow so that failures are handled in a controlled and predictable way.
Solution: Send a notification with relevant error details.
Does the solution meet the goal?
- A. Yes
- B. No
Show answer and explanation ▾
Correct answer: A
Sending a notification that includes the relevant error details directly satisfies both stated requirements. The notification ensures the failure is surfaced to people rather than ending silently, and including the error information, such as the failed action, status and message returned by the external service, lets support staff understand and act on the problem. Placed on the failure path of the flow, the notification step becomes part of a controlled and predictable response to failures or timeouts of the external call, which is exactly what the goal describes.
Why the other options are wrong:
- B. A notification carrying the error details is precisely what the requirements ask for, so it does address the need to avoid silent termination and to report the failure.
Question 8
Note: This question is part of a series of questions that present the same scenario. Each question in the series contains a unique solution that might meet the stated goals. Some question sets might have more than one correct solution, while others might not have a correct solution.
After you answer a question in this section, you will NOT be able to return to it. As a result, these questions will not appear on the review screen.
A company is configuring an agent in Copilot Studio that uses a generative answers node inside multiple topics.
The company requires responses to meet the following requirements:
Must use an executive summary format.
Must apply only within a specific topic.
Must continue using configured knowledge sources.
Must comply with connector security controls.
You need to configure a custom prompt to meet the formatting requirements.
Solution: Remove the generative answers node and rely only on the custom prompt to produce formatted responses.
Does the solution meet the goal?
- A. Yes
- B. No
Show answer and explanation ▾
Correct answer: B
Removing the generative answers node breaks the requirement to keep using the configured knowledge sources. The generative answers node is what searches those sources and grounds the response in their content, and it is also where the response can be shaped for a specific topic. A custom prompt on its own only sends instructions and inputs to a model, so without the node the answers would no longer be grounded in the agent's configured knowledge sources. The correct approach keeps the node in the specific topic and adjusts its prompt to produce the executive summary format.
Why the other options are wrong:
- A. Relying only on a custom prompt drops the grounding in configured knowledge that the generative answers node provides, so the requirement to keep using the configured knowledge sources would not be met.
Question 9
A company uses multiple Copilot Studio agents that perform specialized tasks.
The company needs to enable one agent to directly delegate and invoke another agent within Copilot Studio by using a standard Copilot Studio interface.
You need to configure collaboration among agents by using Copilot Studio.
What should you do?
- A. Add the target agent as a Microsoft Fabric Data Agent.
- B. Provide A2A endpoint information for the target agent.
- C. Upload documentation about the target agent for referencing.
- D. Add the target agent as an indexed knowledge source.
Show answer and explanation ▾
Correct answer: B
The Agent2Agent (A2A) protocol is a standard interface for agents to delegate tasks to one another. Copilot Studio lets you add another agent over A2A by providing the target agent's endpoint information, after which the calling agent can send it requests and use its responses within the conversation. This gives direct delegation and invocation of the specialized agent instead of only reading information about it, which is what the company needs to enable collaboration among its agents.
Why the other options are wrong:
- A. A Microsoft Fabric Data Agent answers questions over Fabric data; it is not a way to register an arbitrary specialized Copilot Studio agent for delegation.
- C. Uploading documentation only lets the agent answer questions about the target agent; it cannot call or delegate work to it.
- D. An indexed knowledge source retrieves stored content for grounding and cannot invoke another agent to perform a task.
Question 10
A company uses an agent in Copilot Studio that produces structured responses using a reusable instruction template. The responses follow the required tone and format. During a compliance review, the team discovers that the responses were generated using a model that is not on the approved list.
The company requires that the reusable instruction template continues to be used and the model used for prompt-driven responses meets governance requirements.
The solution must ensure that prompt-driven responses are generated using an approved model.
You need to configure the solution that meets the requirements.
Which two actions should you perform? Each correct answer presents part of the solution.
NOTE: Each correct selection is worth one point.
- A. Change the prompt to use an approved model, and then publish the agent.
- B. Enable citations so generated responses can be traced and comply with governance requirements.
- C. Add a line to the instruction template that requires the agent to use only approved models when generating responses.
- D. Configure Azure AI Search as the grounding source so the approved model requirement is enforced during response generation.
- E. Ensure the reusable instruction template is the one applied to the agent that generates the structured response.
Show answer and explanation ▾
Correct answer: A, E
Prompt-driven responses run on the model selected for the prompt in the prompt builder, and changes reach users only through the published agent. Changing the prompt to an approved model and then publishing the agent updates the version that users interact with, so runtime responses stop using the unapproved model. Ensuring that the reusable instruction template is the one applied to the agent that generates the structured response keeps the required tone and format, satisfying the requirement that the template continues to be used. Together, these actions deliver structured responses from the approved template on an approved model.
Why the other options are wrong:
- B. Citations show where grounded content came from; they do not change or govern which model generates the response.
- C. Instruction text is read by the model already running the prompt, so a line asking for approved models cannot switch the model that is used.
- D. Azure AI Search only supplies grounding content for answers and has no role in selecting or enforcing the generation model.
Question 11
A company uses an agent in Copilot Studio to coordinate user interactions.
The company wants the agent to delegate requests to a Microsoft Fabric Data Agent that accesses governed business data.
The agent must retrieve structured analytical data while enforcing data security and scope controls.
You need to select the integration approach that meets the requirements.
What should you use?
- A. A Microsoft Fabric Data Agent and bind it to the model.
- B. A REST API action that queries Fabric tables directly.
- C. A Power Automate flow and call it from a topic.
- D. A Copilot connector knowledge source that indexes Fabric data.
Show answer and explanation ▾
Correct answer: A
A Microsoft Fabric Data Agent answers questions over governed Fabric data, such as lakehouses, warehouses and semantic models, and it respects the permissions and scope defined in Fabric. Adding it to the Copilot Studio agent and binding it so the coordinating model can route analytical questions to it lets the agent delegate those requests and receive structured results. Because data access happens through the Fabric Data Agent with the user's permissions, security and scope controls are enforced at the data layer rather than rebuilt in the Copilot Studio agent.
Why the other options are wrong:
- B. A REST API action querying Fabric tables directly bypasses the Fabric Data Agent and its governed scope, and it requires custom query and security handling.
- C. A Power Automate flow called from a topic is not delegation to a Fabric Data Agent and would need its own logic to query and secure the data.
- D. A Copilot connector knowledge source indexes content for search, which does not provide live structured analytics or the Fabric Data Agent's governed access.
Question 12
A company is configuring generative answers for an agent in Copilot Studio.
The agent must use Azure AI Search to retrieve enterprise content, and all responses must be generated by a Microsoft Foundry model.
The agent sometimes generates answers that are not grounded in Azure AI Search content. The company requires that the agent only generate responses when relevant indexed content is retrieved.
You need to configure the agent so that its generative answer behavior meets the organization's requirements.
What should you do?
- A. Use Azure AI Search as a fallback source and enable answer expansion.
- B. Upload documents into the agent and disable fallback responses.
- C. Require grounding for generative answers and use Azure AI Search as the retrieval source.
- D. Enable citations and configure a public model endpoint.
- E. Enable conversational boosting and select a Foundry-approved model.
Show answer and explanation ▾
Correct answer: C
The company needs Azure AI Search to supply the content and wants answers produced only when relevant indexed content is retrieved. Adding Azure AI Search as the agent's knowledge for retrieval keeps the index as the place answers come from, while the Foundry model generates the response text. Requiring grounding, done by turning off Allow ungrounded responses in the Knowledge section of the Generative AI settings, blocks any response produced in a turn where the agent used no knowledge or tool, and the fallback topic triggers instead. Together these settings stop the model from answering from general knowledge when the index returns nothing relevant.
Why the other options are wrong:
- A. Using Azure AI Search only as a fallback with answer expansion still lets other paths answer first and does nothing to block ungrounded responses.
- B. Uploading documents into the agent replaces the required Azure AI Search index with uploaded files, which contradicts the requirement to retrieve content through Azure AI Search.
- D. Citations show where content came from but do not stop ungrounded answers, and a public model endpoint has no bearing on grounding.
- E. Conversational boosting is the system topic that runs generative answers, and choosing a model does not require that responses be grounded in retrieved content.
Question 13
A company uses an agent in Copilot Studio to generate structured responses for internal users.
The agent must meet the following requirements:
Consistently follow a reusable instruction template for response tone and structure.
Use an enterprise-approved foundation model for responses generated through the template.
You need to configure custom prompts so the agent uses the Microsoft Foundry model catalog.
Which two actions should you perform? Each correct answer presents part of the solution.
NOTE: Each correct selection is worth one point.
- A. Publish the agent to apply the Foundry model selection automatically.
- B. Configure Azure AI Search as the grounding source, so the model choice is enforced.
- C. Allow topic authors to define tone and structure directly in topic steps.
- D. Select a model from the Foundry model catalog for the custom prompt configuration.
- E. Create a custom prompt template and apply it where the agent generates the structured response.
- F. Enable citations, so the model is restricted to approved outputs.
Show answer and explanation ▾
Correct answer: D, E
A custom prompt (prompt tool) in Copilot Studio packages reusable instructions, so creating a prompt template that defines tone and structure and applying it at the step that produces the structured response gives consistent output every time. The prompt configuration also has a model selector, and choosing a model from the Microsoft Foundry model catalog there means every response generated through that template uses the enterprise-approved model. Together these two actions cover both requirements: the template controls format, and the selected Foundry model controls which foundation model generates the text.
Why the other options are wrong:
- A. Publishing makes the current agent available on channels; it does not pick a model for a prompt, so the Foundry model must still be selected explicitly.
- B. Azure AI Search supplies grounding content for retrieval and has no effect on which model a custom prompt uses.
- C. Letting each topic author define tone in topic steps produces inconsistent results and does not provide a reusable template or a model choice.
- F. Citations show the sources behind an answer; they do not restrict outputs or select an approved model.
Question 14
An agent in Copilot Studio must complete tasks by automating interactions with a legacy desktop application that does not use APIs.
During testing, the builder must confirm that the automation runs in a supported execution context and can inspect how each automation step executes during conversations.
To meet the business needs, the solution must:
Execute UI automation in a supported, platform-managed environment.
Provide visibility into how automation steps run during agent conversations.
You need to configure the agent based on the requirements.
Which two actions should you perform? Each correct answer presents part of the solution.
NOTE: Each correct selection is worth one point.
- A. Publish the agent so automation settings are applied at runtime.
- B. Inspect automation execution details by using the agent runtime view.
- C. Enable computer use for the agent so UI automation runs in the supported execution environment.
- D. Add the desktop application as a knowledge source so the agent can reference its UI structure.
- E. Configure a custom connector to simulate UI interactions through REST calls.
Show answer and explanation ▾
Correct answer: B, C
Computer use is the Copilot Studio tool that operates websites and desktop apps through a virtual mouse and keyboard, so it can automate a legacy application that has no API. Enabling it for the agent runs the UI automation on a machine set up for computer use in a supported execution environment. For visibility, the builder inspects the execution details at runtime: the test experience shows a step-by-step log of the tool's reasoning and actions next to a preview of the machine, and in conversations the tool shares reasoning messages and images of the machine's activity. These two actions meet the execution and visibility requirements.
Why the other options are wrong:
- A. Publishing releases the agent to channels but does not configure where UI automation runs or show how each automation step executed.
- D. A knowledge source supplies content for answers; it cannot operate a desktop application or describe its UI for automation.
- E. A custom connector calls an API, and the legacy application exposes no API, so REST calls cannot drive its user interface.
Question 15
An agent in Copilot Studio must retrieve real-time data from an external system that exposes a REST API.
The API returns structured JSON data and requires authentication.
The agent must call the API during conversations to fulfill user requests.
The agent must retrieve real-time data from the external system by calling an authenticated REST API during a conversation.
The solution must use a supported mechanism that executes the REST API call and the call must be explicitly configured to run at runtime.
You need to determine which approaches add the REST API to the agent.
Which two solutions meet the goal? Each correct answer is a complete solution.
NOTE: Each correct selection is worth one point.
- A. Configure an HTTP action and invoke it from a topic during the conversation.
- B. Create a topic variable and use a Formula to parse the API endpoint string.
- C. Add the REST API as a tool and call it either by using a topic or agent instruction.
- D. Add the REST API as a knowledge source so the agent can query it during conversations.
- E. Enable generative answers and rely on the model to fetch the REST API data automatically.
- F. Create a global variable to store the API URL and call it using a Redirect node.
Show answer and explanation ▾
Correct answer: A, C
Both approaches execute a real call to the REST API at runtime. An HTTP request action placed in a topic sends the authenticated request when the conversation reaches that node and stores the JSON response in a variable for later nodes. Adding the REST API as a tool, built from its OpenAPI description with authentication configured, lets the agent call it either from a topic node or through generative orchestration guided by the agent's instructions. Each is a complete, explicitly configured mechanism that retrieves live data from the external system during the conversation.
Why the other options are wrong:
- B. A topic variable with a Power Fx formula only manipulates text; parsing an endpoint string never sends a request to the API.
- D. Knowledge sources supply content for generative answers; a REST API is not added as a knowledge source to execute authenticated calls.
- E. Generative answers do not call arbitrary APIs on their own; the call must be configured as a tool or HTTP request.
- F. A Redirect node switches to another topic; storing a URL in a global variable and redirecting does not call the API.
Question 16
A company must enable an agent in Copilot Studio to execute operations exposed by an external MCP server.
The MCP server publishes callable tools and requires authentication.
The agent must be able to invoke MCP tools during conversations.
You need to configure MCP tools so the agent can execute MCP tool calls.
What should you do?
- A. Upload documentation about the MCP server so the agent can answer questions about it.
- B. Enable generative answers so the model can automatically invoke MCP tools.
- C. Publish the agent and rely on default settings for tool availability.
- D. Add the MCP server to the agent's tool configuration.
Show answer and explanation ▾
Correct answer: D
Copilot Studio agents use Model Context Protocol servers through the tools configuration. Adding the MCP server as a tool connects the agent to the server, including the authentication the server requires, and makes the tools the server publishes available to the agent. The agent can then select and call those tools during conversations, with inputs and outputs handled through the protocol. Because the server's tools are discovered from the server itself, they stay current as the server changes, which is the supported way to let the agent execute MCP tool calls.
Why the other options are wrong:
- A. Documentation about the MCP server only lets the agent describe it; the agent still has no connection to call the server's tools.
- B. Generative answers produce grounded responses from knowledge; they do not register or authenticate an MCP server, so no tools become callable.
- C. Publishing with default settings adds nothing; the MCP server must be added and authenticated as a tool before its tools are available.
Question 17
Note: This question is part of a series of questions that present the same scenario. Each question in the series contains a unique solution that might meet the stated goals. Some question sets might have more than one correct solution, while others might not have a correct solution.
After you answer a question in this section, you will NOT be able to return to it. As a result, these questions will not appear on the review screen.
A company is configuring an agent in Copilot Studio that uses a generative answers node inside multiple topics.
The company requires responses to meet the following requirements:
Must use an executive summary format.
Must apply only within a specific topic.
Must continue using configured knowledge sources.
Must comply with connector security controls.
You need to configure a custom prompt to meet the formatting requirements.
Solution: Create a custom prompt tool in the specific topic that instructs the model to respond using an executive-summary structure.
Does the solution meet the goal?
- A. Yes
- B. No
Show answer and explanation ▾
Correct answer: A
This solution meets the goal. A custom prompt tool used inside the specific topic tells the model to structure its output as an executive summary, so the formatting applies only where that topic runs and other topics keep their behavior. The prompt changes how the response is written, not where content comes from, so the configured knowledge sources are still used. Connector security controls such as data policies remain in force because the prompt does not alter connections or bypass governance.
Why the other options are wrong:
- B. A topic-scoped custom prompt with executive summary instructions satisfies the format, scope, knowledge and security requirements together.
Question 18
A team has an agent in Copilot Studio that handles user intake. For some requests, the team wants the agent to delegate work to a Microsoft Foundry agent that performs specialized processing. The Foundry agent already exists.
To meet the business needs, the solution must meet the following requirements:
The primary agent must be able to call the Foundry agent.
The external agent connection must reference the correct Foundry project and agent.
You need to integrate the primary agent with the Foundry agent.
Which two actions should you take? Each correct answer presents part of the solution.
NOTE: Each correct selection is worth one point.
- A. Add an external agent.
- B. Configure the Copilot Studio endpoint URL.
- C. Configure the project endpoint URL.
- D. Add an indexed knowledge source.
- E. Publish the Copilot Studio agent.
Show answer and explanation ▾
Correct answer: A, C
To delegate work to an existing Microsoft Foundry agent, you add it to the primary agent as an external agent, choosing Microsoft Foundry under Connect to an external agent. This makes the Foundry agent callable, and its description helps the primary agent decide when to use it. While adding it, you create the connection to Foundry by configuring the project endpoint URL, which points the integration at the correct Foundry project, and you then identify the specific agent in that project. Adding the external agent meets the calling requirement, and the project endpoint URL meets the requirement to reference the right project and agent.
Why the other options are wrong:
- B. The Copilot Studio endpoint describes how others reach this agent; the connection to Foundry needs the Foundry project endpoint instead.
- D. An indexed knowledge source searches stored content and cannot delegate live processing requests to the Foundry agent.
- E. Publishing the Copilot Studio agent releases it to channels but does not create the external agent link or reference the Foundry project.
Question 19
A company must enable an agent in Copilot Studio to execute operations exposed by an external MCP server.
The MCP server publishes callable tools and requires authentication.
The agent must be able to invoke MCP tools during conversations.
You need to configure MCP tools so the agent can execute MCP tool calls.
What should you do?
- A. Enable generative answers so the model can automatically invoke MCP tools.
- B. Add a Power Platform connector as real-time knowledge to replace MCP tools.
- C. Add the MCP server output as an indexed knowledge source.
- D. Configure authentication for the MCP server connection used by the agent.
Show answer and explanation ▾
Correct answer: D
When an MCP server requires authentication, the agent can only invoke its tools after the connection is set up with the authentication the server expects. In the Copilot Studio MCP onboarding wizard the builder enters the server details and chooses API key (sent in a header or query parameter) or OAuth 2.0, then creates the connection and adds the server to the agent. With that authenticated connection in place, the agent's orchestrator can call the published MCP tools during conversations.
Why the other options are wrong:
- A. Generative answers produce responses from knowledge; they do not authenticate to an MCP server or enable its tools.
- B. Real-time knowledge through a connector only reads data and would not expose the MCP server's callable tools.
- C. Indexing MCP output as knowledge gives read-only content and cannot execute the server's operations.
Question 20
Note: This question is part of a series of questions that present the same scenario. Each question in the series contains a unique solution that might meet the stated goals. Some question sets might have more than one correct solution, while others might not have a correct solution.
After you answer a question in this section, you will NOT be able to return to it. As a result, these questions will not appear on the review screen.
A company is configuring an agent in Copilot Studio that uses a generative answers node inside multiple topics.
The company requires responses to meet the following requirements:
Must use an executive summary format.
Must apply only within a specific topic.
Must continue using configured knowledge sources.
Must comply with connector security controls.
You need to configure a custom prompt to meet the formatting requirements.
Solution: Add instructions in the custom prompt tool to restrict connector access during runtime.
Does the solution meet the goal?
- A. Yes
- B. No
Show answer and explanation ▾
Correct answer: B
This solution does not meet the goal. The requirement is to produce an executive summary format within one topic, and instructions that try to restrict connector access say nothing about response format. Prompt instructions also are not a security control: connector governance is enforced by Power Platform data policies (DLP) and connection settings, not by text in a prompt. The configured knowledge sources and connector controls should simply remain in place while the prompt handles formatting.
Why the other options are wrong:
- A. Restricting connector access through prompt instructions neither applies the executive summary format nor enforces security, which data policies handle.
That was 20 of 116.
The full Microsoft AB-620 pack has all 116 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. $39.00 USD, paid once, with free monthly updates and a pass-or-your-money-back guarantee.
Get the full packNot ready yet? Get the first 58 free as one PDF by email. Get the Half Pack →