Test your AB-620 readiness across every official skill domain. These original questions cover planning and configuring agents, integrating Copilot Studio with enterprise systems, and testing and managing agent solutions. Each answer includes an explanation and a Microsoft Learn source for focused revision.
Q001 - Question
Fabrikam's order management agent in Copilot Studio must work with a shipment tracking agent that Northwind Traders built on its own AI platform. The Northwind agent reasons about shipment context, uses its own logistics tools, and returns contextual responses. A team member suggests calling a tracking API instead. Why is an Agent2Agent (A2A) integration the better fit?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. A2A lets the Fabrikam orchestrator select and invoke each Northwind logistics tool directly.
- B. A2A returns raw shipment data that the Fabrikam agent can format without partner involvement.
- C. A2A delegates tasks to an external agent that keeps its own reasoning, tools, and orchestration across platform and organizational boundaries.
- D. A2A requires Northwind Traders to rebuild its agent in Copilot Studio so that both agents share one platform.
C is correct.
Explanation: The Northwind agent manages its own reasoning, tools, and orchestration. A standard API call isn't sufficient for that kind of agent. A2A is an open standard that lets one agent delegate tasks to another agent across platform and organizational boundaries.
A is incorrect: With A2A, the orchestrator doesn't select the partner's individual tools. The external agent's reasoning and tools stay under its own control.
B is incorrect: Returning raw data describes an API call. The Northwind agent returns contextual, agent-generated responses.
D is incorrect: A2A is designed for agents that run on different platforms. The partner doesn't need to rebuild its agent in Copilot Studio.
Q002 - Question
A company builds a Copilot Studio orchestrator for inventory questions. The architects require the orchestrator to choose which individual inventory tools to call for each request and to combine the results itself. The tools must not be hidden behind another agent's reasoning. Which integration approach should you use?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. Use MCP so that the orchestrator controls tool selection and synthesizes the results.
- B. Use an A2A connection to delegate each inventory request to an external agent.
- C. Use an HTTP connector that calls a single API and returns raw data.
- D. Package the tools as a separate agent on another platform and connect it by using A2A.
A is correct.
Explanation: MCP provides controlled access to tools and data. The orchestrator selects individual tools and synthesizes the results, which matches the requirement.
B is incorrect: A2A delegates a task to an external agent whose reasoning, tools, and orchestration are opaque. The orchestrator can't choose the individual tools.
C is incorrect: An HTTP connector calls an API and returns raw data. It doesn't provide orchestrator-controlled selection across multiple tools.
D is incorrect: Placing the tools behind an external A2A agent hides them behind that agent's reasoning, which conflicts with the requirement.
Q003 - Question
You add an Agent2Agent connection to a Copilot Studio orchestrator. After you enter the partner's endpoint URL, Copilot Studio doesn't populate the agent name and description. Which two actions should you take? Each correct answer presents part of the solution.
Domain: Test and manage agents (20–25%) Type: Multiple choice
- A. Change the authentication type to None so that Copilot Studio can retrieve the agent card.
- B. Verify the endpoint URL and the agent card JSON at /.well-known/agent.json.
- C. Replace the A2A connection with an HTTP connector until the partner publishes a new endpoint.
- D. Enter the name and description manually while the external owner corrects the missing or malformed agent card.
B and D are correct.
Explanation: Copilot Studio retrieves {endpoint}/.well-known/agent.json and populates the name and description only when the agent card is valid. If discovery fails, verify the endpoint and the card JSON. You can enter the name and description manually while the external owner fixes the card.
A is incorrect: Authentication doesn't fix a missing or malformed agent card. Use None only for unprotected development scenarios.
C is incorrect: An HTTP connector returns raw API data instead of an agent-generated response. It doesn't resolve an agent-card discovery failure.
Q004 - Question
A partner organization protects its A2A-enabled agent. The partner gives you a client ID, a client secret, an authorization URL, a token URL, and a refresh URL. You configure the Agent2Agent connection in Copilot Studio. Which authentication option should you select?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. None, because the partner provides an endpoint URL.
- B. OAuth 2.0
- C. API key, with the client secret entered as the key value.
- D. None, with the client secret added to the agent routing description.
B is correct.
Explanation: OAuth 2.0 authentication requires a client ID, client secret, authorization URL, token URL, and refresh URL. These are the values that the partner provided.
A is incorrect: Use None only for unprotected development scenarios. The partner agent is protected.
C is incorrect: API-key authentication requires a header or query-parameter name and a key value. It doesn't use the OAuth 2.0 URLs and client credentials.
D is incorrect: Treat client secrets as sensitive credentials. The routing description is used to route requests, not to store secrets.
Q005 - Question
You connect a Copilot Studio orchestrator to an external shipment tracking agent by using A2A. The orchestrator also has internal agents. Before production, you need to validate that requests route to the correct agent. What should you do?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. Test only in-scope shipment prompts, because out-of-scope prompts are handled by the external agent.
- B. Confirm that the agent card loads, and then publish the orchestrator without further testing.
- C. Broaden the external agent's routing description so that it captures more types of requests.
- D. Use the test canvas and activity log with in-scope, out-of-scope, edge-case, and multi-turn prompts, and then refine the descriptions to resolve routing ambiguity.
D is correct.
Explanation: Test delegation in the Copilot Studio test canvas and activity log with varied prompts. Include in-scope, out-of-scope, edge-case, and multi-turn prompts. Refine the descriptions to resolve routing ambiguity.
A is incorrect: Testing only in-scope prompts doesn't verify that unrelated requests stay away from the external agent.
B is incorrect: A valid agent card confirms discovery only. It doesn't validate routing accuracy or response quality.
C is incorrect: Routing descriptions must be specific and distinct from other agents. A broader description can overlap with internal agents and cause misrouting.
Q006 - Question
Contoso's HR agent has a topic that routes employees by leave category: Annual leave, Sick leave, or Parental leave. The HR team requires that the conversation does not continue until the employee picks one of these three recognized options. What should you configure?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. Quick replies of type Send a message, one for each leave category, below the message node
- B. Message variations that list the three leave categories in different phrasings
- C. A Question node with multiple-choice options for the three leave categories
- D. Quick replies of type Send a hidden message, one for each leave category
C is correct.
Explanation: A Question node enforces a choice. It doesn't advance the conversation until the employee selects or enters a recognized option. Use a Question node when you need to constrain input.
A is incorrect: Quick replies are suggestions, not requirements. Employees can ignore them and type freely, and some channels don't display quick replies at all.
B is incorrect: Message variations only rotate alternate phrasings of the same message. They don't collect or enforce a selection.
D is incorrect: A hidden message quick reply is still a quick reply. It guides the employee but doesn't prevent free-text input.
Q007 - Question
You need the Contoso HR agent to show each employee a leave balance summary with entitlement total, days used, and remaining days side by side. The values differ for each employee and are calculated earlier in the topic. The card must only display data and must not collect input. What should you configure?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. An Adaptive Card in a Send a message node that uses a ColumnSet for the three values and references topic variables with Power Fx
- B. An Ask with Adaptive Card node that uses Input.Number elements for the three values
- C. An Adaptive Card in a Send a message node with the three values hardcoded in TextBlock elements
- D. A message node with a bulleted list and quick replies for each value
A is correct.
Explanation: Informational Adaptive Cards appear in Send a message nodes and display content without collecting input. Use a ColumnSet to show comparable values side by side. Compute the runtime values in topic variables, then reference them in card properties with Power Fx.
B is incorrect: Ask with Adaptive Card nodes collect user input. This requirement is to display data, not to have employees enter numbers.
C is incorrect: Hardcoded values can't reflect each employee's balance. Bind topic variables with Power Fx instead.
D is incorrect: Message node formatting improves readability, but it doesn't present structured data in a side-by-side visual layout. Quick replies also don't display data values.
Q008 - Question
The Contoso HR agent uses an Ask with Adaptive Card node for leave requests. You find two issues. Some employees close the card without submitting, and later nodes fail. After a retry, some employees select Submit on an earlier card that is still in the conversation. Which two actions should you take? Each correct answer presents part of the solution.
Domain: Plan and configure agent solutions (30–35%) Type: Multi-select
- A. Mark the leave type and reason fields as Required in the properties panel
- B. Add a Condition node immediately after the Ask with Adaptive Card node to check that required variables, such as leave type, aren't blank
- C. Change the Action.Submit Title from Submit to Submit request
- D. Include a unique identifier in each submit action's data payload and validate it when you process the response
B and D are correct.
Explanation: If an employee closes the card without selecting Submit, the output variables remain empty. A Condition node routes empty values to a handling branch. When multiple cards are sent in a conversation, a unique identifier in each submit action's data payload lets you verify which card and action the response came from.
A is incorrect: Required prevents submission without values. It doesn't create any values when the employee closes the card without submitting.
C is incorrect: The Title only changes the button label. It doesn't distinguish responses from earlier cards or handle dismissed cards.
Q009 - Question
Contoso expands its leave request form to several regions. Valid leave categories differ by region and are retrieved from the HR system by an HTTP request earlier in the topic. Employees must select from the valid categories in the Ask with Adaptive Card form. What should you do?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. Replace the dropdown with an Input.Text field so employees can type their leave category
- B. Keep hardcoded choices in the card designer and add every category from every region
- C. Add quick replies for each leave category below the card
- D. Bind the Input.ChoiceSet choices property to a Power Fx expression that references a topic variable table with title and value columns
D is correct.
Explanation: Power Fx lets you bind a topic variable to an Input.ChoiceSet choices property by using a = expression in the card JSON. The topic variable must be a table with title and value columns. Use this pattern when the option list varies by user, region, or policy.
A is incorrect: Input.Text accepts open-ended text. Use Input.ChoiceSet when the employee should select from a predefined set of options.
B is incorrect: Hardcoded choices suit stable lists. They would show categories that aren't valid for the employee's region.
C is incorrect: Quick replies are suggestions that employees can ignore, and they don't populate the card's input variables.
Q010 - Question
The Contoso HR agent is deployed to Microsoft Teams and to the company's internal web portal. A maker copied a leave summary card from the Adaptive Cards Hub. The card declares version 1.6 and renders correctly in the Copilot Studio test pane. HR wants consistent rendering in both channels with low maintenance. What should you do?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. Keep version 1.6 because the card renders correctly in the Copilot Studio test pane
- B. Set the card version to 1.5, use only v1.5 schema elements, then publish and test the card in Teams and the web portal
- C. Create a v1.6 card and a v1.5 card, and use topic-level conditions to route each channel to a different card
- D. Keep version 1.6 and replace Action.Submit with Action.Execute for the card buttons
B is correct.
Explanation: Microsoft Teams supports Adaptive Card schema v1.5, and the web channel supports v1.6. Designing to v1.5 gives consistent rendering across both channels without special handling. Verify the version field on copied cards, and test in each actual target channel.
A is incorrect: The test pane uses the web channel engine, which supports v1.6. A v1.6 card can look correct there and still fail or fall back to plain text in Teams.
C is incorrect: Channel detection with card variants works, but you must keep two card definitions in sync. That added maintenance isn't justified for this requirement.
D is incorrect: The web channel doesn't support Action.Execute. Use Action.Submit, and a v1.6 card still has a compatibility risk in Teams.
Q011 - Question
Woodgrove Bank wants its IT service desk agent in Copilot Studio to submit requests to an internal IT service management API. No prebuilt connector exists for this API. The organization also plans to use the same API integration in Power Automate and Power Apps. What should you use?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. A REST API tool created in Copilot Studio from an OpenAPI v2 specification
- B. A connector added to the agent as a knowledge source
- C. A custom connector that exposes the API through OpenAPI, added to the agent as a connector tool
- D. A prebuilt standard connector selected from the connector gallery
C is correct.
Explanation: A custom connector exposes an organization API through OpenAPI. You can reuse it across Copilot Studio, Power Automate, Power Apps, and Azure Logic Apps, which meets the reuse requirement.
A is incorrect: A REST API tool is created in Copilot Studio for a single use case when no connector exists. It is also preview-only and not intended for production use.
B is incorrect: A connector used as a knowledge source queries live data to ground responses. It does not let the agent submit requests in an external system.
D is incorrect: No prebuilt connector exists for this internal API.
Q012 - Question
A developer at Woodgrove Bank creates a custom connector for the CMDB API. Another maker must add this custom connector as a tool to the IT service desk agent in Copilot Studio. What should you do before the maker adds the tool?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. Share the custom connector with the maker by using Can view or Can edit permission in the Power Apps maker portal
- B. Share the connection with the maker by using Can use + share permission in Power Apps Connections
- C. Recreate the CMDB integration as a REST API tool in the agent
- D. Assign Power Fx formulas to all inputs of the connector action
A is correct.
Explanation: Share a custom connector with the intended makers by using Can view or Can edit permission in the Power Apps maker portal before they add it as a tool.
B is incorrect: Sharing a connection with Can use + share permission lets other makers manage and test that connection. It does not share the custom connector itself.
C is incorrect: A REST API tool is for cases where no connector exists. It is also preview-only and not suitable for production use.
D is incorrect: Input configuration defines how values are supplied to the action. It does not give the maker access to the custom connector.
Q013 - Question
In a Woodgrove Bank escalation conversation, the agent must retrieve a CMDB configuration record at a specific step every time. The call must use fixed parameter values. The team does not want the generative orchestrator to decide whether to call the action. What should you configure?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. An agent-level connector tool with a more detailed description
- B. The CMDB connector as a knowledge source for the agent
- C. Connector tool inputs set to dynamically fill with AI
- D. A topic-level call to the connector action from a topic node
D is correct.
Explanation: Topic-level connector calls give deterministic control over when the action runs and which parameters it uses. Use a topic node when execution must be predictable.
A is incorrect: Agent-level tools are selected by generative orchestration. A better description improves selection, but the orchestrator still decides when to invoke the tool.
B is incorrect: A connector used as a knowledge source grounds responses with live data. It does not provide deterministic action execution at a specific step.
C is incorrect: Inputs that dynamically fill with AI do not meet the requirement for fixed parameter values.
Q014 - Question
Woodgrove Bank adds a connector tool that looks up employee IT profiles. Each employee has an account in the target system. Each employee must see only the data that their own account permits. Which credential mode should you use for the connector tool?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. Maker-provided credentials
- B. User-provided credentials
- C. A REST API tool with the None authentication option
- D. A connection shared with all employees by using Can use + share permission
B is correct.
Explanation: User-provided credentials are the default mode. Each user signs in at first tool use, and the agent acts with that user's permissions. Use this mode when data access must be scoped to each user.
A is incorrect: Maker-provided credentials use one shared identity for all users. This suits a common data scope, not per-user access, and it requires an authenticated channel.
C is incorrect: None authentication sends calls without a user identity. It cannot scope data to each employee's permissions.
D is incorrect: Can use + share permission on a connection lets other makers manage and test it. It does not scope data access for each end user.
Q015 - Question
You create a REST API tool in Copilot Studio from an OpenAPI specification. The target API requires an API key that is sent in a request header named x-api-key. You select API key authentication. Which two settings must you configure? Each correct answer presents part of the solution.
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Multi-select
- A. The actual parameter name that the API expects
- B. The identity provider client credentials and token endpoint
- C. Header as the parameter location
- D. A connection shared with other makers by using Can use + share permission
A and C are correct.
Explanation: API key authentication for a REST API tool requires a display label, the actual parameter name, and a location of Header or Query. For this API, set the parameter name to x-api-key and set the location to Header.
B is incorrect: Identity provider credentials and endpoint configuration are required for OAuth 2.0, not for API key authentication.
D is incorrect: Sharing a connection with Can use + share permission applies to connector tool connections in Power Apps. It is not part of API key configuration for a REST API tool.
Q016 - Question
Woodgrove Bank has a premium wealth management topic in Microsoft Copilot Studio. The topic contains a generative answers node that uses only a premium-tier SharePoint site as its node-level source. Compliance requires answers to come only from approved premium content. The topic must always give the user a useful next step, even when no relevant content is found. What should you do?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. Add AI general knowledge as a node-level source so the node always generates an answer.
- B. Rely on the agent-level knowledge sources to answer when the node-level source returns no content.
- C. Add a Condition node after the generative answers node to handle the no-answer case with a fallback message or escalation path.
- D. Add more SharePoint sites as node-level sources to increase the chance that the node finds content.
C is correct.
Explanation: The evidence states that if no relevant content is found, the generative answers node returns a "no information found" response. To ensure a useful response, add a Condition node after the generative answers node to handle the no-answer case with a fallback message or escalation path.
A is incorrect: Adding AI general knowledge uses the model's built-in knowledge, which conflicts with the compliance requirement to use only approved premium content.
B is incorrect: Agent-level sources act as a fallback only when a node has no sources of its own. Because this node has a node-level source, the agent-level sources are not used.
D is incorrect: Adding more node-level sources returns broader results and dilutes relevance, and it does not guarantee a useful response when no content is found.
Q017 - Question
A maker wants a generative answers node to use an existing Azure AI Search index that is connected through Azure OpenAI Service. The maker opens the Data source pane on the Create generative answers node. The option isn't in the standard knowledge sources panel. What should the maker do?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. Upload the indexed files to Dataverse and select Documents as the node-level source.
- B. Select Classic data, and then select Azure OpenAI on your data.
- C. Select Bing Custom Search and point it at the Azure AI Search index.
- D. Add the index as a SharePoint URL so the node retrieves results through Microsoft Graph Search.
B is correct.
Explanation: The evidence states that Azure OpenAI on your data connects the node to an Azure AI Search index through Azure OpenAI Service. Because this option is not in the standard knowledge sources panel, you must select Classic data in the Data source pane to access it.
A is incorrect: The Documents option searches files uploaded to Dataverse. It does not connect the node to an existing Azure AI Search index.
C is incorrect: Bing Custom Search uses a configured Bing Custom Search instance. It is not the correct source type for an Azure AI Search index.
D is incorrect: The SharePoint source requires a SharePoint URL and Microsoft Entra ID authentication. It does not connect to an Azure AI Search index.
Q018 - Question
Woodgrove Bank uses an HTTP request node to call a regulatory API. A Set variable node maps the response to a Power Fx Table with Content, ContentLocation, and Title columns, and the Table is passed to the Custom data field of a generative answers node. The API returns about 50 regulations sorted by publication date. Users report that answers often miss the regulation that matches their question. What should you do?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. Pass the user's question as a query parameter so the API filters and ranks results and returns the most relevant records first.
- B. Increase the number of records the API returns so the node has more content to rank by relevance.
- C. Replace the HTTP request with a Power Automate flow that returns the full set of regulations to the node.
- D. Remove the ContentLocation and Title columns so the node processes only the Content values.
A is correct.
Explanation: The evidence states that only the first three records in the Table are used to generate the response. Because the selection is positional and not relevance-based, you must design your data retrieval to filter and rank results before returning them so the three most relevant records reach the node.
B is incorrect: Additional records beyond the first three are ignored by the node. Increasing the number of records does not help the node rank by relevance.
C is incorrect: While a Power Automate flow is useful for complex retrieval logic, the three-record limit still applies. Returning the full set of regulations does not resolve the ranking issue.
D is incorrect: ContentLocation and Title are optional columns used for citations. Removing them does not change which records the node processes for the answer.
Q019 - Question
Woodgrove Bank's agent has an investment product topic and a support topic. Both topics use generative answers nodes and share the same knowledge sources. The investment topic must return formal responses that use bullet points for multi-step answers. The support topic must use an empathetic tone. Which two actions should you take? Each correct answer presents part of the solution.
Domain: Plan and configure agent solutions (30–35%) Type: Multi-select
- A. Create a second agent for the support topic because custom instructions apply to every topic in an agent.
- B. Enter the formal tone and bullet point instructions in the custom instruction field of the investment topic's generative answers node.
- C. Configure separate knowledge sources for each topic because custom instructions apply to a knowledge source.
- D. Enter the empathetic tone instructions in the custom instruction field of the support topic's generative answers node.
B and D are correct.
Explanation: The evidence states that custom instructions apply only to the generative answers node where you configure them. You can configure formal instructions in the investment topic's node and empathetic instructions in the support topic's node. Both topics can share the same agent and knowledge sources while keeping their instructions independent.
A is incorrect: Custom instructions do not affect other topics in the agent, so creating a second agent is not required.
C is incorrect: Custom instructions are configured on a generative answers node, not on a knowledge source. Both topics can use the same knowledge sources.
Q020 - Question
Woodgrove Bank built a custom prompt in Prompt builder to classify documents that clients submit. The team connected a model deployment from Microsoft Foundry to the prompt and tested it with representative and edge-case samples. The team now plans to use the prompt in production. What should the team do before production use?
Domain: Plan and configure agent solutions (30–35%) Type: Single choice
- A. Switch the prompt back to GPT-4.1 mini because production prompts must use the default model.
- B. Move the Foundry deployment to the generative answers node's Data source pane so the node can classify documents.
- C. Remove the prompt inputs so the model classifies documents without topic variables.
- D. Review the deployment costs and ensure Power Platform DLP governance for the connection.
D is correct.
Explanation: The evidence states that when you connect a Microsoft Foundry deployment in Prompt builder, you must review deployment costs and ensure Power Platform DLP governance before production use.
A is incorrect: While GPT-4.1 mini is the default model, you can select other models. You should test and use the model that fits the required quality for the task.
B is incorrect: Generative answers retrieves and synthesizes configured knowledge. Document classification is a structured reasoning task that requires a custom prompt.
C is incorrect: You must map topic variables to the prompt inputs. Removing the inputs prevents the prompt from receiving the document it needs to classify.
Q021 - Question
A bank has several Copilot Studio agents, including an account-opening agent. Each agent needs tools from the same group of backend systems. A central platform team wants to publish these tools once, maintain them in one place, and make updates available to connected agents without republishing each agent. What should you recommend?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. Add a separate REST API tool for each backend system to every agent.
- B. Publish the tools on a centrally managed MCP server and connect each agent to that server.
- C. Create a separate connector tool for each backend system in every agent.
- D. Create topics in each agent that call each backend system individually.
B is correct.
Explanation: An MCP server publishes tools once for all connected agents. Updates made on the server become available to those agents without republishing them. This approach avoids separate integrations with each backend system.
A is incorrect: You define REST API tools separately in each agent. This approach repeats the integration work for each backend system and each agent.
C is incorrect: You define connector tools separately in each agent. This approach does not give you one centrally published set of tools.
D is incorrect: Topics in each agent that call each backend system create repeated integrations. This approach does not use a centrally managed tool source.
Q022 - Question
A maker is building a Copilot Studio agent for a single department. At a specific step in a topic, the agent must call a backend operation. What should the maker use for this integration?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. Connect an MCP server and call the MCP tool directly from the topic.
- B. Turn off generative orchestration and connect an MCP server.
- C. Edit the MCP tool description in Copilot Studio so that the topic can select the tool.
- D. Use a connector or REST API tool in the agent.
D is correct.
Explanation: A topic-triggered call is a use case for a connector or a REST API tool. You define these tools in the agent. They give you more control than MCP tools and work for one-off integrations.
A is incorrect: Topics cannot call MCP tools directly. Only generative orchestration can invoke MCP tools.
B is incorrect: MCP tools require generative orchestration. If you turn off generative orchestration, the agent cannot invoke MCP tools.
C is incorrect: You cannot edit MCP tool descriptions in Copilot Studio. Also, a description change does not let a topic call an MCP tool.
Q023 - Question
A partner hosts a custom MCP server that uses Server-Sent Events (SSE) as its transport. You need to connect a Copilot Studio agent to this server. What should you do first?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. Ask the server owner to migrate the server to Streamable HTTP, and then enter the HTTPS Streamable HTTP endpoint in the onboarding wizard.
- B. Select SSE as the transport type in the MCP onboarding wizard.
- C. Add the partner server from the MCP catalog so that Microsoft manages the endpoint configuration.
- D. Configure OAuth 2.0 authentication so that the agent can connect to the SSE endpoint.
A is correct.
Explanation: Copilot Studio supports only the Streamable HTTP transport. The server URL must be the HTTPS Streamable HTTP endpoint that the server owner provides. An SSE server must migrate before you can connect to it.
B is incorrect: SSE has not been supported since August 2025. You cannot use SSE to connect the server.
C is incorrect: The MCP catalog provides prebuilt Microsoft connectors. A custom partner server requires onboarding configuration.
D is incorrect: The authentication type does not change the transport. The server must still use Streamable HTTP.
Q024 - Question
You connect an agent to an MCP server that returns sensitive production data. The server supports OAuth 2.0, but its identity provider doesn't support Dynamic Client Registration (DCR). Which OAuth 2.0 option should you configure?
Domain: Test and manage agents (20–25%) Type: Single choice
- A. Dynamic discovery
- B. Dynamic
- C. Manual
- D. None
C is correct.
Explanation: Use Manual when the identity provider doesn't support DCR. Provide the client credentials and configure the endpoints. If the identity provider requires it, register the generated callback URL.
A is incorrect: Dynamic discovery requires DCR with discovery endpoints. The identity provider doesn't support DCR.
B is incorrect: Dynamic requires DCR. With this option, you provide the authorization and token URLs because discovery isn't available.
D is incorrect: None is suitable for trusted, network-controlled servers. Use API key or OAuth 2.0 for sensitive production data.
Q025 - Question
An account-opening agent is connected to a shared MCP server that exposes many tools. Allow all is on. During testing, the orchestrator sometimes selects tools that the account-opening workflow doesn't need. Some server-published tool descriptions are vague. Which two actions should you take? Each correct answer presents part of the solution.
Domain: Test and manage agents (20–25%) Type: Multiple choice
- A. Edit the vague MCP tool descriptions on the Tools page in Copilot Studio.
- B. Turn off Allow all and enable only the tools that the account-opening workflow needs.
- C. Create topics that call the correct MCP tools directly.
- D. Ask the MCP server owner to update the vague tool descriptions on the server.
B and D are correct.
Explanation: When you turn off Allow all and enable only the tools that the workflow needs, you reduce orchestration ambiguity and apply least privilege. The orchestrator uses server-published tool descriptions to select tools. Only the server owner can update those descriptions.
A is incorrect: MCP tool descriptions are read-only in Copilot Studio. The server owner must update inadequate descriptions.
C is incorrect: Topics cannot call MCP tools directly. Only generative orchestration can invoke MCP tools.
Q026 - Question
Adatum Corporation has a customer service agent in Copilot Studio. The team wants to add a tool that the AI orchestrator can select automatically during any conversation. The team doesn't need the tool to run at a fixed step in a specific topic. Where should the team add the tool?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. Add the tool inside one topic by using Add node and then Add a tool.
- B. Add a Send HTTP request node to each topic that might need the tool.
- C. Add the tool at the agent level.
- D. Call an agent flow from a condition node in a single topic.
C is correct.
Explanation: The evidence states that tools added at the agent level are selected by the AI orchestrator automatically during any conversation. Use this approach when the tool doesn't need to run at a specific, controlled point in a topic.
A is incorrect: A tool added inside a topic runs at a specific point on the topic canvas. It isn't selected by the orchestrator across all conversations.
B is incorrect: The Send HTTP request node is an ad-hoc REST API call within a topic. It runs where you place it and doesn't let the orchestrator select it during any conversation.
D is incorrect: Calling an agent flow from a topic runs the flow at that step in that topic only. It doesn't meet the requirement for orchestrator selection across conversations.
Q027 - Question
A maker adds a connector tool to a topic in Copilot Studio. The connector reads records from a business system. Each user must only get data that their own permissions allow. Which authentication option should the maker configure for the connector tool?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. Maker-provided authentication, so the tool runs with the maker's identity.
- B. End-user authentication.
- C. Store the maker's credentials in an environment variable and reference it in the connector tool.
- D. Replace the connector tool with a Send HTTP request node that uses a single static API key.
B is correct.
Explanation: The evidence states that end-user authentication uses each user's permissions when the connector tool runs. This meets the requirement that users only get data their own permissions allow.
A is incorrect: Maker-provided authentication uses a shared maker identity for all users. It also requires an authenticated channel.
C is incorrect: This approach still uses one shared identity for every user. It doesn't apply each user's permissions.
D is incorrect: A single static API key is one shared credential, not each user's permissions. The HTTP request node is also intended for cases where no connector exists.
Q028 - Question
A maker wants to call an existing agent flow from a topic by using Add a tool. The flow doesn't appear in the tool picker in Copilot Studio. Which two conditions could cause this issue? Each correct answer presents a complete solution. Select two.
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Multiple choice
- A. The flow doesn't use the When an agent calls the flow trigger.
- B. The Asynchronous response toggle in the Respond to the agent action is set to Off.
- C. The flow includes actions after the Respond to the agent action.
- D. The flow isn't a solution flow.
A and D are correct.
Explanation: The evidence states that the "When an agent calls the flow" trigger makes a flow callable from a topic or the agent orchestrator. A flow without this trigger doesn't appear in the tool picker. The flow must also be a solution flow to appear in Copilot Studio.
B is incorrect: Setting Asynchronous response to Off is a requirement. It makes the topic wait for the flow outputs before it continues.
C is incorrect: Actions after the Respond to the agent action are allowed. They run asynchronously after the topic receives its outputs.
Q029 - Question
A customer-facing topic uses a Send HTTP request node to call a carrier tracking API. When the API is unavailable, the conversation stops without a useful message. The maker wants the topic to show a clear fallback message and not expose raw error details. What should the maker configure?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. Keep the default Raise an error mode so the system On Error topic handles the failure.
- B. Increase the Request timeout value so the request doesn't fail.
- C. Change the body option to Raw content so errors are returned as text in the chat.
- D. Select Continue on error, save the status code and error response to topic variables, and add a condition that checks the status code.
D is correct.
Explanation: The evidence states that configuring "Continue on error" lets the topic continue after the node. You save the HTTP status code and error response to variables, such as Topic.StatusCode and Topic.ErrorResponse, and then add a condition. A value of 200 shows the result, and other values route to a fallback message.
A is incorrect: "Raise an error" triggers the system On Error topic and stops the current flow. This might be acceptable for internal agents, but not for a customer-facing topic.
B is incorrect: A longer Request timeout can help when an API responds slowly under load. It doesn't handle an API that is unavailable or a tracking number that doesn't exist.
C is incorrect: Raw content controls the request body format sent to the API. It doesn't handle failed responses, and showing raw error details in the chat isn't the goal.
Q030 - Question
A maker configures a Send HTTP request node with a GET request to a carrier tracking API. The request succeeds. The maker now needs a message node to show the shipment status field from the JSON response. What should the maker do?
Domain: Integrate and extend agents in Copilot Studio (40–45%) Type: Single choice
- A. Set Response data type to From Sample Data, use Get schema from sample JSON with a sample response, save the response to Topic.TrackingResponse, and reference Topic.TrackingResponse.status.
- B. Change the method to POST and select JSON Content for the request body.
- C. Add a Content-Type: application/json header to the request.
- D. Move the request into an agent flow and return the status by using the Respond to the agent action.
A is correct.
Explanation: The evidence states that you must define the response shape so Copilot Studio can expose the response fields as variables. You paste a sample JSON response from the API documentation, confirm the schema, and save the response to a variable. Then use Power Fx dot notation, such as Topic.TrackingResponse.status.
B is incorrect: Retrieving shipment status uses GET with No Content. Changing the method and body doesn't define the response schema.
C is incorrect: The Content-Type header tells the API how to parse a request body. It doesn't expose response fields as topic variables.
D is incorrect: An agent flow is for multi-step workflows or reusable logic. This integration is a single request in one topic, so the HTTP request node response schema is enough.