Creating connections programmatically between Search AI and Azure Foundry Project

[MicroMichaelIE]

Is there way to add connection to the AI Foundry Project? For a Foundry "Hub" we can use the MLClient class in the python sdk. There's documentation on how to create connections through the Azure portal for AI Foundry Project, but not how to do it with code.

I am yet to find a way to create a connection to my AI Search resource programmatically, either by SDK, REST API, or even terraform. If this feature doesn't exist, what is the way to do this as it's unrealistic to expect half of the workflow to be in code and the other half still waiting for manual action. If it does exist could someone point me in the correct direction?

[leestott]

Hi @MicroMichaelIE

So, YES there is a way to programmatically create connections to Azure AI Search within an AI Foundry Project.

Programmatic Connection Options

Python SDK via AIProjectClient
You can use the AIProjectClient from azure.ai.projects to list and reference existing connections. However, creating new connections (e.g. to Azure AI Search) currently requires either:

• Pre-creating the connection in the portal or CLI
• Using the [azapi Terraform provider](https://registry.terraform.io/providers/Azure/azapi/latest/docs) to invoke lower-level ARM APIs

Terraform with azapi_resource
You can create connections to Azure AI Search using Terraform and the azapi_resource block. Here's a simplified example:

resource "azapi_resource" "ai_search_connection" {
  type = "Microsoft.MachineLearningServices/workspaces/connections@2025-01-01-preview"
  name = "my-search-connection"
  parent_id = azurerm_ai_foundry_project.example.id
  location = "global"

  body = {
    properties = {
      category = "AzureAISearch"
      target = var.search_endpoint
      authType = "ApiKey"
      isSharedToAll = true
      metadata = {
        ApiType = "Azure"
        ResourceId = var.search_resource_id
      }
      credentials = {
        key = var.search_api_key
      }
    }
  }
}

This approach is used in [Carlos Mendible’s Terraform walkthrough](https://carlos.mendible.com/2025/05/20/deploy-azure-foundry-and-bing-grounding-with-terraform/) and aligns with the [official Foundry connection schema](https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/how-to/configure-project-connection).

REST API
The Foundry SDK blog confirms that a unified REST API exists for managing connections, To use this today you need to use the Microsoft.MachineLearningServices/workspaces/connections endpoint directly with an authenticated ARM client.

How to do this

• Use the [AI Foundry Connections GitHub repo](https://github.com/Azure-Samples/AI-Foundry-Connections) for Python samples and best practices.
• If you're Terraform-first, follow the [official guide for hub creation and connection provisioning](https://learn.microsoft.com/en-us/azure/ai-foundry/how-to/create-hub-terraform).
• For SDK workflows, you can list and consume connections via project_client.connections.list() — but creation still requires CLI or azapi.

[amynic]

@MicroMichaelIE if @leestott answer helps please 'Mark as answer'