azure-foundry-agent

Azure Foundry Agent Skill

Invoke Azure Foundry agents using the Azure MCP tools. This skill enables discovering Foundry resources, listing agents, and invoking agents for task classification and other AI workflows.

Configuration

Copy config.example.yaml to config.yaml and customize:

# Azure subscription (optional - uses current if not set)
subscription: ""

# Foundry resource name (from foundry_resource_get)
resource: "shboyer-copilot-proj-resource"

# Project name
project: "shboyer-copilot-proj"

Default agent

agent: "priority-detection-agent"

Keywords that trigger this skill

triggers:

• "rate my task"
• "prioritize"

The endpoint is constructed automatically:
`https://<resource>.services.ai.azure.com/api/projects/<project>`

When a user message contains any trigger keyword, use this skill's workflow.

## Prerequisites

- Azure MCP tools must be available (`azure-mcp-foundry`)
- User must be authenticated to Azure (`az login`)
- A Foundry project with deployed agents

## Workflow

### Step 1: Get Current Subscription

First, determine the user's current Azure subscription:

```bash
az account show --query "{subscriptionId:id, name:name}" -o json

Step 2: List Foundry Resources

Use the Azure MCP to list Foundry resources in the subscription:

Tool: azure-mcp-foundry
Command: foundry_resource_get
Parameters:
  subscription: <subscription-id-from-step-1>

This returns all AIServices resources with their endpoints and deployed models.

Step 3: List Agents in a Project

Once you have the resource endpoint, list available agents:

Tool: azure-mcp-foundry
Command: foundry_agents_list
Parameters:
  endpoint: https://<resource-name>.services.ai.azure.com/api/projects/<project-name>

Note: The endpoint format for agents uses .services.ai.azure.com (not .cognitiveservices.azure.com).

Step 4: Invoke an Agent

Connect to and query a specific agent:

Tool: azure-mcp-foundry
Command: foundry_agents_connect
Parameters:
  agent-id: <agent-id-from-step-3>
  endpoint: https://<resource-name>.services.ai.azure.com/api/projects/<project-name>
  query: <user-message>

Example: Task Priority Classification

For the priority-detection-agent:

Input: "Create a new PO for new supplies"

Output:

{
  "priority": "MEDIUM",
  "reasoning": "Creating a purchase order for new supplies is an important task that supports ongoing operations. While it should be addressed promptly, it is typically part of planned work and does not constitute an emergency.",
  "task_content": "Create a new PO for new supplies"
}

Priority Guidelines

• HIGH: Critical emergencies - production outages, security breaches, system failures, customer-impacting issues
• MEDIUM: Important tasks with reasonable deadlines, planned work, non-critical bugs
• LOW: Nice-to-have improvements, cosmetic changes, tasks marked "when time permits"

Available Commands Reference

Command	Purpose	Required Parameters
foundry_resource_get	List Foundry resources	subscription
foundry_agents_list	List agents in a project	endpoint
foundry_agents_connect	Invoke an agent	agent-id, endpoint, query
foundry_agents_create	Create a new agent	See MCP docs
foundry_threads_list	List conversation threads	endpoint
foundry_threads_get-messages	Get thread messages	endpoint, thread-id

Error Handling

• Missing subscription: Run az account show to get current subscription
• Authentication failed: Run az login to authenticate
• Agent not found: Use foundry_agents_list to verify agent ID
• Invalid endpoint: Ensure endpoint uses .services.ai.azure.com/api/projects/ format

Tips

1. Cache the subscription ID and endpoint for subsequent calls
2. The agent ID from foundry_agents_list (e.g., asst_xxx) is used in foundry_agents_connect
3. Thread IDs from agent responses can be used to continue conversations
4. Use foundry_resource_get without resource-name to list all resources