How can we retrieve daily MCP tool invocation counts for Copilot Studio agents?

Enable environment-level telemetry to Application Insights, then query ExecuteTool spans in the dependencies table for daily MCP invocation counts, agent, tool, outcome, timestamp, and environment ID.^1^

Configure telemetry export

Environment-level telemetry is currently preview, applies only to managed environments, and is the appropriate scope for centralized tool-execution monitoring across agents. It supports both the standard and Microsoft Copilot harnesses, although Application Insights logs are available only for standard-harness agents built in Copilot Studio.^1^

Prerequisites include:

• Power Platform Administrator or Dynamics 365 Administrator at tenant level.
• Environment Administrator or System Administrator in Dataverse.
• Contributor, Writer, or Administrator rights to the Application Insights resource.
• An Application Insights resource dedicated to the environment or tenant.
• Local authentication enabled for the Application Insights integration.^2^

Configure it as follows:

1. Open Power Platform admin center → Manage → Data export.
2. Select App Insights → New data export.
3. Select Copilot Studio as the export type.
4. Select the source environment.
5. Select the Azure subscription, resource group, and Application Insights resource.
6. Create the export package.
7. Run a test conversation that invokes an MCP tool.
8. Allow up to 24 hours for initial telemetry delivery, then open Application Insights → Logs and verify that ExecuteTool rows appear in dependencies.^2^

Query daily MCP tool usage

Copilot Studio exports InvokeAgent, ExecuteTool, and OutputMessages as dependency spans. Tool spans include:

• timestamp
• success and resultCode
• gen_ai.agent.id and gen_ai.agent.name
• gen_ai.tool.name
• gen_ai.tool.type
• gen_ai.tool.call.id
• env.id
• Optional arguments, results, and failure information^1^

Run this in Application Insights → Logs:

dependencies
| where timestamp >= startofday(ago(30d))
| where name == "ExecuteTool"
| extend
    AgentId = tostring(customDimensions["gen_ai.agent.id"]),
    AgentName = tostring(customDimensions["gen_ai.agent.name"]),
    ToolName = tostring(customDimensions["gen_ai.tool.name"]),
    ToolType = tostring(customDimensions["gen_ai.tool.type"]),
    ToolCallId = tostring(customDimensions["gen_ai.tool.call.id"]),
    EnvironmentId = tostring(customDimensions["env.id"])
| where ToolType startswith "MCP"
| summarize
    TotalCalls = count(),
    SuccessfulCalls = countif(success == true),
    FailedCalls = countif(success == false)
  by Day = startofday(timestamp), EnvironmentId,
     AgentId, AgentName, ToolType, ToolName
| order by Day desc, AgentName asc, ToolName asc

For an invocation-level audit rather than daily aggregation:

dependencies
| where timestamp >= ago(30d) and name == "ExecuteTool"
| extend
    AgentName = tostring(customDimensions["gen_ai.agent.name"]),
    ToolName = tostring(customDimensions["gen_ai.tool.name"]),
    ToolType = tostring(customDimensions["gen_ai.tool.type"]),
    ToolCallId = tostring(customDimensions["gen_ai.tool.call.id"]),
    EnvironmentId = tostring(customDimensions["env.id"]),
    Failure = tostring(customDimensions["Status.message"])
| where ToolType startswith "MCP"
| project timestamp, EnvironmentId, AgentName, ToolName,
          ToolType, ToolCallId, success, resultCode, Failure
| order by timestamp desc

Because telemetry properties can evolve during preview, inspect the live schema before operationalizing a dashboard:

dependencies
| where timestamp > ago(7d)
| mv-expand Key = bag_keys(customDimensions) to typeof(string)
| summarize Events = make_set(name),
    SampleValue = take_any(tostring(customDimensions[Key])) by Key
| order by Key asc

Mapping to the requested metrics

gen_ai.tool.name can contain a server-qualified value—for example, workiqsharepoint:mcp_SharePointRemoteServer—but that naming convention should not be assumed universally. Check the values emitted by each MCP integration. If they consistently encode server identity, extract it with a deployment-specific parsing rule; otherwise, correlate with telemetry emitted by the MCP server itself.

For Business Central specifically, its server-side MCP telemetry provides configurationName, configurationSystemId, toolName, dynamicToolName, and invocation outcome. That schema is Business Central-specific and should not be generalized to every Copilot Studio MCP server.^3^

Historical and API access

Historical analysis is available only for telemetry collected after the export is enabled; the export does not document retroactive backfill. Azure Monitor Logs supports KQL analysis and configurable retention. Application Insights Analytics retention is normally 90 days and can be extended, with total retention available up to 12 years depending on configuration.^4^

There is no documented standalone Copilot Studio API dedicated to historical MCP invocation counts. For programmatic access, query the exported Application Insights/Azure Monitor data using the Application Insights Query API or Azure Monitor Logs Query API.^5^

The Application Insights → Agents (preview) → Tools view also aggregates ExecuteTool spans by call frequency and performance, but KQL is required for the requested environment, agent, tool, date, and outcome breakdown.^1^

References

1. Environment-level telemetry with Application Insights (preview)
2. Export data to Application Insights
3. Analyze MCP server tool calls telemetry
4. Azure Monitor Logs overview
5. Azure Monitor REST API index