Edit

Share via

Quickstart: Use SQL MCP Server with Azure AI Foundry

Important

SQL MCP Server is in preview and this documentation and the engine implementation is subject to change during this evaluation period.

This quickstart shows you how to connect a deployed SQL MCP Server to Azure AI Foundry using a Custom MCP Tool. You test everything using the Chat Playground with a chat model such as GPT-5-mini - no coding required.

Note

The new Foundry experience is in preview. You need to select the preview toggle in the header when you access https://ai.azure.com.

Diagram that shows a deployed SQL MCP Server connected to Azure AI Foundry.

Prerequisites

Deployed SQL MCP Server

You need a SQL MCP Server deployed to Azure Container Apps with public ingress enabled. If you don't have a deployment yet, complete the Deploy SQL MCP Server to Azure Container Apps quickstart first.

You need the MCP endpoint URL from that deployment (for example, https://your-app.azurecontainerapps.io/mcp).

Azure AI Foundry project

You need access to Azure AI Foundry with an existing project and access to a chat model (for example, GPT-5-mini).

Step 1: Access Azure AI Foundry

  1. Navigate to https://ai.azure.com.
  2. In the header, select the new Foundry experience (preview toggle).
  3. Select your Foundry project.
  4. In the left navigation, select Playground.

Note

UI text and navigation may vary as Azure AI Foundry evolves. For the latest guidance, see MCP tools in Azure AI Foundry.

Step 2: Create or open an agent

  1. Select an existing agent or create a new one.
  2. In the agent configuration, locate the Tools section on the left panel.

Step 3: Add a Model Context Protocol tool

  1. In the Tools section, select Add (the button may show a dropdown or + icon).
  2. Select Add a new tool from the dropdown.
  3. The Select a tool dialog opens with three tabs: Configured, Catalog, and Custom.
  4. Select the Custom tab at the top.
  5. Select Model Context Protocol (MCP) from the available options.
  6. Select Create to proceed.

Step 4: Configure your MCP connection

The Add Model Context Protocol tool dialog appears. Fill in the following information:

Name (required)

Enter a descriptive name: products-mcp.

Remote MCP Server endpoint (required)

Enter your MCP server URL: https://<your-container-app-url>/mcp (from your Azure Container Apps deployment).

Authentication (required)

Select Unauthenticated from the dropdown.

Note

This configuration works because the Azure Container Apps quickstart configured anonymous permissions (for example, anonymous:read). At the time of writing, this quickstart uses Unauthenticated mode. If you enable authentication on your MCP server, configure the MCP tool accordingly (authentication configuration is not covered in this quickstart).

Select Connect to add the tool.

Step 5: Configure agent instructions

In the Instructions section at the top of the agent configuration, add or replace the text with:

You are a helpful product catalog assistant. When answering questions about products, use the products-mcp tool to query the database. 

The Products entity contains:
- Id: Product identifier
- Name: Product name
- Inventory: Units in stock
- Price: Retail price in USD
- Cost: Store cost in USD

Always use the schema discovery tool first to understand the schema, then use the query tool to retrieve data.

Step 6: Test your agent

Try these prompts in the chat:

Example 1: Schema discovery

What tables or entities are available in the database?

The agent should call a schema discovery tool (such as describe_entities) to see the Products entity structure.

Example 2: Simple query

List all products

The agent should call a query tool (such as read_records) to retrieve product data.

Example 3: Price query

Show me products under $20

The agent should query products where Price < 20.

Example 4: Inventory check

Which products are low in stock (less than 30 units)?

The agent should filter Inventory < 30.

Example 5: Complex query

What's the most expensive product, and how many do we have in stock?

The agent should:

  • Query products sorted by Price descending
  • Return the top result with Inventory count

Step 7: View tool calls

In the Chat Playground, you can see tool calls and their arguments/responses:

  • Tool called: The schema discovery or query tool used
  • Arguments: The parameters passed (filters, fields, etc.)
  • Response: The data returned from your SQL database

Troubleshooting

Tool not appearing in Azure AI Foundry

  • Verify the MCP URL is correct and accessible.
  • Check that the Container App is running with public ingress enabled (see monitoring guidance).
  • Test the /mcp endpoint using curl or a REST client to verify reachability.

Tool calls require approval

  • In the MCP tool configuration, ensure Require approval is set to never.

Agent not using the tool

  • Check your system message includes clear instructions about when to use the tool.
  • Try asking more specific questions that relate to the Products data.

For deployment issues, see the troubleshooting section in the Azure Container Apps deployment guide.

Related content

  • Last updated on