📋 Shared Instructions: shared-instructions.md - Cross-cutting concerns.

Add Microsoft Copilot Studio

Workflow

Check Memory Bank → 2. Add Connector → 3. Configure → 4. Build → 5. Update Memory Bank

Step 1: Check Memory Bank

Check for memory-bank.md per shared-instructions.md.

Step 2: Add Connector

First, find the connection ID (see connector-reference.md):

Run the /list-connections skill. Find the Microsoft Copilot Studio connection in the output. If none exists, direct the user to create one using the environment-specific Connections URL — construct it from the active environment ID in context (from power.config.json or a prior step): https://make.powerapps.com/environments/<environment-id>/connections → + New connection → search for the connector → Create.

pwsh -NoProfile -Command "pac code add-data-source -a microsoftcopilotstudio -c <connection-id>"

Step 3: Configure

Ask the user which Copilot Studio agent they want to invoke and what operations they need.

Agent Setup Prerequisites (manual steps the user must complete in Copilot Studio):

Publish the agent: In Copilot Studio, click Channels → select Teams → add to Teams → click Publish.
Get the agent name: Under Channels, click "Web app". The connection string URL contains the agent name. Example: https://...api.powerplatform.com/copilotstudio/dataverse-backed/authenticated/bots/cr3e1_myAgent/conversations?... — the agent name is cr3e1_myAgent.

ExecuteCopilotAsyncV2 -- execute an agent and wait for the response:

Use the ExecuteCopilotAsyncV2 operation (path: /proactivecopilot/executeAsyncV2). This is the only endpoint that reliably returns agent responses synchronously. It is the same endpoint used by Power Automate's "Execute Agent and wait" action.

const result = await MicrosoftCopilotStudioService.ExecuteCopilotAsyncV2({
  message: "Your prompt or data here", // Can be a JSON string
  notificationUrl: "https://notificationurlplaceholder" // Required by API but unused; any URL works
});

// Response structure:
// result.responses — Array of response strings from the agent
// result.conversationId — The conversation ID
// result.lastResponse — The last response from the agent
// result.completed — Boolean indicating if the agent finished

Important: Agents often return responses as JSON strings. Parse the responses array to extract meaningful data:

const agentResponse = result.responses?.[0];
if (agentResponse) {
  const parsed = JSON.parse(agentResponse);
  // Extract specific fields, e.g., parsed.trend_summary
}

Use Grep to find specific methods in the generated service file (generated files can be very large — see connector-reference.md).

Known Issues

ExecuteCopilot (/execute) -- fire-and-forget, only returns ConversationId, not the actual response. Do NOT use this.
ExecuteCopilotAsync (/executeAsync) -- returns 502 "Cannot read server response" errors. Do NOT use this.
Conversation turn model (/conversations/{ConversationId}) -- only works after /execute, which doesn't provide responses. Do NOT use this.
Response casing varies -- check all variations: conversationId, ConversationId, conversationID.

Step 4: Build

npm run build

Fix TypeScript errors before proceeding. Do NOT deploy yet.

Step 5: Update Memory Bank

Update memory-bank.md with: connector added, agent name configured, configured operations, build status.

add-mcscopilot