Microsoft Foundry MAI Speech MCP connector for Power Platform

April 06, 2026

MAI-Transcribe-1 launched April 2, 2026, as part of Microsoft’s three-model MAI announcement alongside MAI-Voice-1 and MAI-Image-2. It delivers state-of-the-art speech-to-text accuracy at 3.9% average word error rate, 2.5x faster batch transcription than Azure Fast, and ranks first on the FLEURS benchmark in 11 core languages across the top 25 most-used languages.

Built for messy, real-world audio environments—not clean studio recordings. Meeting rooms with background chatter, phone calls with connection noise, field recordings with wind and traffic.

This connector brings MAI-Transcribe-1 and Whisper into Power Platform with two MCP tools for Copilot Studio and three REST operations for Power Automate and Power Apps. A borrowed chat completion operation lets agents summarize, analyze, or discuss transcription results within a conversation.

Full source: GitHub repository

The models

Both models support the same operations—transcription and translation—through separate REST endpoints that take a deployment ID parameter. Select the model at deployment time.

Tools

MCP tools for Copilot Studio

How it works

User: "Transcribe the recording from today's standup meeting"
      [provides audio URL]

1. Orchestrator calls transcribe_audio({
     audio_url: "https://storage.blob.core.windows.net/.../standup.mp3",
     language: "en",
     response_format: "verbose_json"
   })

   → Returns:
     text: "Good morning everyone. Let's start with updates..."
     language: "en"
     duration: 847.5
     segments: [{ start: 0.0, end: 3.2, text: "Good morning everyone." }, ...]

User: "Summarize that meeting and pull out the action items"

2. Orchestrator calls chat_completion({
     prompt: "Summarize this transcript and list action items:
              Good morning everyone. Let's start with updates...",
     system_prompt: "You are a meeting notes assistant"
   })

   → Returns structured meeting summary with action items

REST operations for Power Automate and Power Apps

Parameter reference

Response format options

Transcription response

Use cases

Meeting transcription and minutes: Transcribe a meeting recording, then use chat_completion to extract action items, decisions, and key discussion points. The verbose_json format provides timestamps for linking notes back to specific moments.

Call recording analysis and compliance: Process customer service or sales calls through TranscribeAudio, then analyze for compliance keywords, sentiment, or escalation triggers using the chat completion tool.

Multilingual content transcription: MAI-Transcribe-1 supports 25+ languages with auto-detection. Transcribe content in the original language or use TranslateAudio to translate any supported language into English text.

Subtitle generation: Use the srt or vtt response format to generate ready-to-use subtitle files for video content. Drop the output directly into your video publishing workflow.

Voice memo processing: Build a Power Automate flow that watches a SharePoint folder for audio uploads, transcribes each file, and stores the text alongside the original recording.

Podcast and media transcription: Process long-form audio content for searchable archives, show notes, or accessibility compliance.

Prerequisites

1. An Azure subscription with access to Microsoft Foundry
2. Deploy a speech-to-text model from the Foundry Model Catalog:
   • MAI-Transcribe-1 — for state-of-the-art transcription accuracy
   • Whisper — for broad language support
3. Note the Resource Name (for example, my-foundry-resource from https://my-foundry-resource.openai.azure.com) and API Key from the deployment

Setting up the connector

1. Deploy a speech model in Microsoft Foundry

1. Go to the Foundry Model Catalog
2. Search for MAI-Transcribe-1 or Whisper and deploy the model
3. Copy the Resource Name and API Key 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 and API Key
3. Select Create connection

4. Test the connector

Test TranscribeAudio by uploading a short audio file (mp3, wav, or m4a) and setting the deployment ID to your model name (for example, MAI-Transcribe-1).

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 “Transcribe this audio file” or “What was discussed in this recording?”

The MCP transcribe_audio tool downloads audio from a URL, so provide publicly accessible links or SAS URLs for files in Azure Storage.

Known limitations

• Audio files must be 25 MB or smaller
• Supported formats: mp3, mp4, mpeg, mpga, m4a, wav, webm
• For files larger than 25 MB, use the Azure Speech batch transcription API instead
• The MCP transcribe_audio tool downloads audio from the provided URL—files must be publicly accessible or use a SAS URL
• The chat_completion tool is borrowed from the parent Microsoft Foundry connector and requires a chat-capable model deployed to the same resource
• The connector uses two different host patterns: .openai.azure.com for audio operations and .services.ai.azure.com for chat and MCP

Files

Resources

• MAI Speech connector source code
• Microsoft Foundry Model Catalog
• Microsoft Foundry API
• MAI-Transcribe-1 model card (PDF)
• 3 new MAI models announcement
• MAI Playground (US only)