Microsoft Foundry MAI Voice MCP connector for Power Platform

April 07, 2026

A flow triggers when a new blog post publishes. It generates an audio narration, uploads the MP3 to SharePoint, and links it in the post metadata. No recording studio, no human narrator, no manual steps.

MAI-Voice-1 launched April 2, 2026, as part of Microsoft’s three-model MAI announcement alongside MAI-Transcribe-1 and MAI-Image-2. It generates 60 seconds of audio in 1 second with emotional range, speaker identity preservation, and custom voice cloning from just a few seconds of sample audio. Already powering Copilot Audio Expressions and Copilot Podcasts.

This connector brings MAI-Voice-1 and Azure Neural TTS voices into Power Platform with three MCP tools for Copilot Studio and three REST operations for Power Automate and Power Apps. A borrowed chat completion tool helps agents generate SSML markup and select voices within a conversation.

Full source: GitHub repository

The model

MAI-Voice-1 delivers top-tier voice generation with nuance, emotional range, and expression that preserves speaker identity across long-form content. Custom voice cloning requires separate enrollment through Azure Speech Studio.

Tools

MCP tools for Copilot Studio

How it works

User: "Read this announcement in a warm, professional voice"

1. Orchestrator calls list_voices({
     locale_filter: "en-US"
   })

   → Returns available voices with styles:
     [
       { shortName: "en-US-JennyNeural", gender: "Female",
         styles: ["cheerful", "sad", "angry", "excited", ...] },
       { shortName: "en-US-GuyNeural", gender: "Male",
         styles: ["newscast", "angry", "cheerful", ...] },
       ...
     ]

2. Orchestrator calls synthesize_speech({
     text: "We're excited to announce...",
     voice: "en-US-JennyNeural",
     language: "en-US",
     output_format: "audio-48khz-192kbitrate-mono-mp3"
   })

   → Returns: { status: "success", audio_size_bytes: 48320,
                 voice: "en-US-JennyNeural" }

User: "Generate SSML with emphasis on the product name"

3. Orchestrator calls chat_completion({
     prompt: "Write SSML that emphasizes 'Power Platform'
              in this sentence: Power Platform now supports
              voice-enabled agents.",
     system_prompt: "You are an SSML expert"
   })

   → Returns SSML with prosody and emphasis tags

REST operations for Power Automate and Power Apps

Parameter reference

Audio output formats

SSML input

The Synthesize Speech operation takes SSML (Speech Synthesis Markup Language) as input, giving you control over voice selection, language, prosody, emphasis, breaks, and speaking styles:

<speak version='1.0' xml:lang='en-US'>
  <voice xml:lang='en-US' name='en-US-JennyNeural'>
    <prosody rate="medium" pitch="default">
      Welcome to today's update.
      <break time="500ms"/>
      Here's what's new.
    </prosody>
  </voice>
</speak>

Use the chat_completion tool or operation to have an LLM generate SSML from plain text with the right voice, pacing, and emphasis for your content.

List Voices response

Each voice includes:

Use cases

Automated narration pipeline: Build a Power Automate flow that watches for new content in SharePoint, synthesizes audio narration, and stores the MP3 alongside the original document. Training materials, policy updates, and announcements get audio versions automatically.

Voice agent responses: Give your Copilot Studio agent a voice. Use synthesize_speech to convert the agent’s text responses to audio for phone-based or accessibility-first scenarios.

Multilingual voice output: List voices filtered by locale, select the right voice for the target language, and synthesize speech—all within the same flow. Support content in dozens of languages without maintaining separate voice configurations.

Podcast and audio content generation: Use high-quality output formats (audio-48khz-192kbitrate-mono-mp3) for professional-grade audio. Combine with MAI-Transcribe-1 for a full audio content pipeline—transcribe interviews, edit the text, re-synthesize as polished narration.

Branded voice experiences: Clone a specific voice through Azure Speech Studio, then use this connector to generate branded audio content at scale. Product announcements, customer communications, and training materials all sound consistent.

Accessibility: Convert text content to speech for users who prefer or need audio. Reports, dashboards, notifications—anything text-based can get an audio version through a simple flow.

Prerequisites

1. An Azure subscription with access to Microsoft Foundry
2. Speech Services enabled on your Foundry resource (MAI-Voice-1 is available through the Foundry Model Catalog)
3. Note the Resource Name, API Key, and Region (for example, eastus2) from the Azure portal

Setting up the connector

1. Enable Speech Services

1. Go to Microsoft Foundry
2. Ensure Speech Services are enabled on your resource
3. Copy the Resource Name, API Key, and Region from the deployment page

2. Create the custom connector

1. Go to Power Platform Maker Portal
2. Navigate to Custom connectors > + New custom connector > Import an OpenAPI file
3. Upload apiDefinition.swagger.json
4. On the Security tab:
   • Authentication type: API Key
   • Parameter label: API Key
   • Parameter name: api-key
   • Parameter location: Header
5. On the Code tab:
   • Enable Code
   • Upload script.csx
6. Select Create connector

3. Create a connection

1. Select Test > + New connection
2. Enter your Resource Name, API Key, and Region
3. Select Create connection

4. Test the connector

Test ListVoices first to see available voices. Then test SynthesizeSpeech with simple SSML:

<speak version='1.0' xml:lang='en-US'>
  <voice xml:lang='en-US' name='en-US-JennyNeural'>
    Hello, this is a test of the MAI Voice connector.
  </voice>
</speak>

5. Add to Copilot Studio

1. In Copilot Studio, open your agent
2. Add this connector as an action—Copilot Studio detects the MCP endpoint via x-ms-agentic-protocol
3. Test with prompts like “List available English voices” or “Convert this text to speech”

Known limitations

• Audio output is capped at 10 minutes per request
• SSML body length is limited by the Speech Services API
• Custom voice cloning requires separate enrollment through Azure Speech Studio
• The MCP synthesize_speech tool generates audio but cannot return binary data directly to the agent—use the REST SynthesizeSpeech operation to retrieve audio files
• The connector uses two different host patterns: {region}.tts.speech.microsoft.com for speech operations and .services.ai.azure.com for chat and MCP
• Connection requires three parameters (Resource Name, API Key, Region) unlike other Foundry connectors that need only two

Files

Resources

• MAI Voice connector source code
• Microsoft Foundry Model Catalog
• MAI-Voice-1 model card (PDF)
• 3 new MAI models announcement
• SSML reference
• Azure Speech Service voices
• MAI Playground (US only)