Design Guide: The Conversation Surface
Alexa+ uses a turn-based conversational model. A customer turn produces a response from Alexa (voice, screen, or both), and your add-on participates one turn at a time.
Voice and screen are rendered together
MCP responses produce a voice response and, where supported, an on-screen response. Alexa incorporates your returned data into the voice response. The customer's utterance and the quality, depth, and breadth of your data shape this spoken reply. You influence Alexa's response through the data you return, not by scripting it directly.
Interruptions and maintaining continuity across turns
After interruptions, whether initiated by the customer or the system, Alexa resumes the conversation where appropriate. For example:
C: "Ok add a double cheeseburger with fr–"
A: <Timer audio and visuals play>
C: "Alexa, stop the timer."
A: <stops timer, re-surfaces previous task on screen>
C: "Alexa, let's add that double cheeseburger with fries, and go ahead and check out"
Your add-on may also be invoked across multiple turns within a single conversation, and customers may move between add-ons or shift topics in the same session. Treat each tool response as self-contained while assuming Alexa will weave it into the broader conversation. For example:
C: "Alexa, I want to finish booking that hotel we were chatting about for our Vegas trip."
A: "Sure, we had you set up with a king suite with a view at Aria, for $479 a night. Is that the one you want to book?"
C: "What's the weather going to be like there?"
A: "Next week in Las Vegas, expect partly sunny skies, with an average high of 76, and a low of 55."
C: "Alright let's book that room then."
A: "Here's your booking details for the king suite with a view at Aria. Your total with taxes and fees is …"
Related topics
| Go back to: Introduction | Next: Tools, Schema, and Data Design |
Last updated: Jul 21, 2026
