What it does

Apps in agents lets an agent bring app capabilities — form completion, data exploration, visualisation, or a request for human input — straight into the conversation, powered by the Power Apps MCP Server. Users act on an app without navigating to it.

Key facts

• Backed by the Power Apps MCP Server; app actions surface as tools the agent can call.
• Example: reading an Outlook email, ask Copilot to update a Power Apps record from the thread in place.
• Announced at Ignite 2025; tied to the new Power Apps and Copilot Chat direction.
• Runs where the user already is — Copilot Chat, Outlook, and other M365 surfaces.
• Governed under the same Power Platform environment and DLP controls as the underlying app.

When to use / skip

Use when you want users to complete an app task inside a conversation instead of context-switching — quick record updates, capturing input, showing data. Skip for heavy, multi-screen app workflows that genuinely need the full app UI.

Configuration decisions

• Which app capabilities to expose to the agent.
• Environment and DLP scope for the app being surfaced.
• How much the agent should do automatically vs prompt for confirmation.

Gotchas

• It's new and evolving — expect capability and naming changes over the coming releases.
• Data actions still run under Power Platform governance; a restrictive DLP policy will block the tool with little explanation to the user.
• In-conversation write-back needs a confirmation step or you'll get unintended edits.

Consultant notes

• This blurs the line between "app" and "agent" — useful, but it means Power Platform ALM and DLP now govern conversational surfaces too.
• Costs follow the Power Platform model (licences, capacity), not Copilot grounding — price it in that lane.
• Good demo material for stakeholders who want to see agents do work, not just answer questions.

---

Review as apps-in-agents moves from announcement to documented GA.