Hi everyone,

I’m testing an agent built in Microsoft Copilot Studio that uses the Microsoft Learn documentation MCP tools to retrieve Microsoft documentation and answer user questions.

I’m facing a strange difference between the Copilot Studio Test Pane and Microsoft Teams.

The issue is:

In Copilot Studio Test Pane:

• I ask a question → I get the correct answer.
• I ask the same question again → I get the answer again.
• I can repeat the question multiple times → The agent continues to respond correctly.
• The Microsoft Learn MCP tool is able to retrieve the relevant documentation.

In Microsoft Teams:

• I ask the question → I get an answer.
• When I ask the same or similar question again, sometimes Teams becomes silent / doesn't return an answer.
• The same question still works correctly in the Copilot Studio Test Pane.

So the behavior looks something like:

>

>

The agent is using Microsoft Learn MCP tools for retrieving Microsoft documentation, so I'm wondering whether the issue could be related to how the MCP tool call, grounding, conversation state, or published agent is handled through the Teams channel.

I have already tried republishing the agent and starting a fresh Teams conversation.

Has anyone experienced this with a Copilot Studio agent + Microsoft Learn MCP Server/tools + Teams?

I'm particularly interested in knowing what I should check in Application Insights / Copilot Studio logs to determine whether:

• The Teams request is reaching the agent.
• The Microsoft Learn MCP tool is being called on subsequent turns.
• The MCP tool returns results but the agent doesn't generate a response.
• Conversation/session state is affecting the second or third question.
• There is a difference between the Test Pane runtime and the Teams published runtime.
• Grounding / "Allow ungrounded responses" is affecting subsequent turns.
• Teams is using an older published version or cached conversation state.

Any suggestions or similar experiences would be really helpful. Thanks!

Hi everyone,

I’m working on an Agent in Microsoft Copilot Studio that uses the Microsoft Learn MCP Server as its tool.

I’m currently facing a significant response latency issue and would like to understand if this is expected and how I can optimize it.

My setup

• Microsoft Copilot Studio Agent
• Microsoft Learn MCP Server
• The agent uses the Microsoft Learn MCP tools to retrieve Microsoft documentation and related information.
• No SharePoint knowledge source or other external knowledge source is involved.

Latency I’m seeing

During testing, I’m seeing approximately:

Scenario	Response time
Simple question that triggers Microsoft Learn MCP	~27–39 seconds
Question that triggers MCP and produces a longer response	~65 seconds

The difference is quite significant because questions that don't invoke the MCP tool respond almost immediately, while questions requiring the Microsoft Learn MCP tool can take 30–60+ seconds.

My questions

1. Is this level of latency expected when using the Microsoft Learn MCP Server with Copilot Studio?
2. Is there a recommended way to reduce the latency of Microsoft Learn MCP tool calls?
3. Could the latency be caused by the MCP server retrieving/fetching documentation before returning the result?
4. Is there a way to optimize the agent so that it retrieves only the minimum required information from Microsoft Learn instead of fetching large amounts of documentation?
5. Does the number of MCP tool calls made during a single user request significantly affect the overall response time?
6. Are there any recommended Copilot Studio configuration or prompt/instruction changes that can reduce the time taken when the Microsoft Learn MCP tool is invoked?
7. For anyone using Copilot Studio + Microsoft Learn MCP, what kind of response times are you seeing, and what optimizations have worked for you?

I’m specifically looking for practical ways to bring the response time down, ideally from 30–60+ seconds to a few seconds for straightforward documentation questions.

Any guidance from people who have implemented this setup would be greatly appreciated.

Thanks!

I’m trying to understand the billing model for Microsoft Copilot Cowork.

Copilot Cowork appears to be a cloud-based SaaS application, and its usage-based billing can be associated with an Azure subscription. However, I’m unclear whether Azure credits provided through an MSDN/Visual Studio subscription can actually be used to cover Copilot Cowork usage charges.

My questions are:

1. Can MSDN/Visual Studio Azure credits be used for MS Copilot Cowork usage?
2. If not, why are these credits ineligible even though Cowork is a cloud-based SaaS application and its billing can be linked to an Azure subscription?
3. Is there a distinction between an Azure subscription being used as a billing vehicle and Azure credits actually being eligible to cover the charges?
4. Has anyone tested this or found official Microsoft documentation that explicitly confirms whether MSDN/Visual Studio Azure credits can or cannot be used for Copilot Cowork?

I’m specifically looking for answers based on Microsoft documentation or actual billing experience.

I’m trying to understand the billing model for Microsoft Copilot Cowork.

Copilot Cowork appears to be a cloud-based SaaS application, and its usage-based billing can be associated with an Azure subscription. However, I’m unclear whether Azure credits provided through an MSDN/Visual Studio subscription can actually be used to cover Copilot Cowork usage charges.

My questions are:

1. Can MSDN/Visual Studio Azure credits be used for MS Copilot Cowork usage?
2. If not, why are these credits ineligible even though Cowork is a cloud-based SaaS application and its billing can be linked to an Azure subscription?
3. Is there a distinction between an Azure subscription being used as a billing vehicle and Azure credits actually being eligible to cover the charges?
4. Has anyone tested this or found official Microsoft documentation that explicitly confirms whether MSDN/Visual Studio Azure credits can or cannot be used for Copilot Cowork?

I’m specifically looking for answers based on Microsoft documentation or actual billing experience.

Hi everyone,

I'm learning how multi-agent systems work in enterprise environments and trying to understand how Agent-to-Agent communication works in practice.

Scenario

I have two agents:

• Agent 1 — CoWork Agent: The main assistant for IT administrators. It understands user requests, plans the workflow, and delegates tasks.
• Agent 2 — ServiceNow Agent: A specialized agent that knows how to interact with ServiceNow APIs — fetching incidents, updating tickets, and resolving issues.

A user asks the CoWork Agent:

"Check all open ServiceNow incidents and resolve any Microsoft 365-related issues automatically."

The CoWork Agent understands the request but doesn't directly interact with ServiceNow — it delegates to the ServiceNow Agent. Here's how I imagine the flow:

1. CoWork Agent → ServiceNow Agent: "Fetch all open incidents."
2. ServiceNow Agent → CoWork Agent: "Found 25 open incidents."
3. CoWork Agent → ServiceNow Agent: "Identify which ones are Microsoft 365-related."
4. ServiceNow Agent → CoWork Agent: "10 incidents match."
5. CoWork Agent → ServiceNow Agent: "Auto-resolve password reset incidents and update the tickets."
6. ServiceNow Agent → CoWork Agent: "7 tickets resolved and updated."
7. CoWork Agent presents the final summary to the user.

My Questions

1. Is this considered true Agent-to-Agent (A2A) communication, or is it just an orchestrator calling a tool?
2. How would the CoWork Agent communicate with the ServiceNow Agent in a real implementation — REST API, message queue, a shared tool registry, or something else?
3. Does the ServiceNow Agent expose tools that the CoWork Agent calls, or are they fully independent agents with their own reasoning?

I'm interested in understanding how real-world enterprise multi-agent systems are designed beyond simple tool calling.

Thanks!

Hi everyone, I’ve been experimenting with integrating Microsoft Copilot Cowork with ServiceNow using plugins and custom skills.

Right now, the setup can respond to questions, retrieve KB articles, and guide users on resolving incidents. But I’m thinking beyond a normal chatbot use case.

The idea is to make it work more like an automation and orchestration layer:

• Detect incoming ServiceNow tickets
• Understand issue context
• Check the appropriate ServiceNow queue (where tickets wait to be picked by teams)
• Search related KB articles
• Generate troubleshooting or resolution steps
• Trigger actions automatically where possible
• Update ticket status and notify users

Example flow:
User raises a common access issue → Ticket enters the ServiceNow support queue → Copilot Cowork identifies the pattern → retrieves relevant KB → suggests or executes the resolution workflow → updates the ticket automatically.

From my understanding, normal Copilot interactions are mostly question-answer based, but I’m curious whether Cowork can become more action-driven and reduce manual ticket handling.

Has anyone tried something similar?
How far can we push automation here, and what limitations or architecture challenges did you find?

Hi everyone,

I am currently exploring a POC where we want to integrate Microsoft Copilot Cowork with ServiceNow.

Scenario:

A user raises a ServiceNow ticket like:

"VPN not getting connected"

Expected flow:

• Copilot/Cowork should understand the issue
• Search ServiceNow KB articles
• Return troubleshooting/resolution steps automatically
• Optionally perform actions if supported

I have some architecture questions:

1. Where does Copilot Cowork actually sit in this architecture?
2. Do we need to create a separate Copilot Studio Agent first and then use Cowork?
3. Does Cowork directly integrate with ServiceNow, or does it go through Copilot Studio / Connectors / APIs?
4. What actions are actually possible after integration?

Examples:

• KB lookup
• Ticket status retrieval
• Incident creation
• Ticket updates
• Trigger workflows
• Escalation
• Service catalog requests

1. Has anyone implemented a similar setup? What architecture worked best?

Looking for guidance from anyone who has worked with Microsoft Copilot + ServiceNow integrations.

Thanks.