Azure MCP Agent (Custom)

This repository contains a custom MCP server (server.py) exposing tools to:

1. List active Azure role assignments for the signed-in user.
2. List Azure resources (subscription-wide or specific resource group).
3. Deploy resources via Bicep-backed PowerShell scripts (storage account, key vault, Azure OpenAI, AI Search, AI Foundry).

Prerequisites

• Windows + PowerShell (pwsh) recommended.
• Python 3.10+ installed.
• Azure CLI installed: https://learn.microsoft.com/cli/azure/install-azure-cli
• Logged in to Azure: az login --use-device-code (or normal az login).
• Appropriate RBAC permissions to list and deploy resources.

Python Setup

cd "c:\Users\v-siddjha\OneDrive - MAQ Software\Desktop\agent"
python -m venv .venv
. .venv\Scripts\Activate.ps1
pip install -r requirements.txt

Run MCP Server Directly

python server.py

This starts the MCP process that tools (e.g., GitHub Copilot Agent) can attach to.

Configure GitHub Copilot to Use This MCP Server

In VS Code settings (JSON view), add an MCP server entry. Example:

"github.copilot.mcpServers": {
  "azure-agent": {
    "command": "python",
    "args": [
      "c:/Users/v-siddjha/OneDrive - MAQ Software/Desktop/agent/server.py"
    ],
    "env": {
      "PYTHONUNBUFFERED": "1"
    }
  }
}

Restart VS Code or Copilot Agent session. Copilot should list available tools (azure_login, list_permissions, list_resources, get_resource_parameters, deploy_resource).

Tool Usage (Within Copilot Chat)

• List permissions:
  • Prompt: Call tool list_permissions (optionally pass user_principal_name or out_file).
• List resources (all):
  • Prompt: Call tool list_resources.
• List resources for RG:
  • Prompt: Call tool list_resources {"resource_group_name":"my-rg"}.
• Discover required params for deployment:
  • Prompt: Call tool get_resource_parameters {"resource_type":"storage-account"}.
• Deploy a resource (must supply required params):
  • Prompt: Call tool deploy_resource {"resource_type":"storage-account","parameters":{"ResourceGroupName":"my-rg","StorageAccountName":"mystorage123","Location":"eastus","AccessTier":"Hot"}}.

Local Test Harness (Without Copilot)

Run the included script:

python test_agent.py

Uncomment the deployment section and fill real parameter values before testing resource creation.

Adding New Deployment Scripts

1. Place new deploy-*.ps1 in scripts/.
2. Add a Bicep template in templates/.
3. Extend DEPLOYMENT_SCRIPTS in server.py with logical key -> script filename.
4. Use get_resource_parameters to see inferred required params.

Notes on Parameter Validation

deploy_resource infers required parameters by detecting if (-not $Param) prompts in the script. Ensure each truly required input has that pattern for stricter enforcement.

Common Issues

• Azure CLI not found: Install CLI and restart shell.
• Not logged in: Run az login.
• Missing permissions: Confirm role assignments (e.g., Reader, Contributor) with list_permissions tool output.
• Deployment failures: Check Bicep template, naming constraints, region availability.

Security Considerations

• Do not hardcode secrets; use Key Vault deployment then store secrets securely.
• Validate user-supplied parameters before passing to scripts if exposing externally.

Cleanup

To remove test resources, use Azure Portal or az group delete -n <RGName> --no-wait --yes.

Happy building! Use the Copilot Agent chat to invoke tools interactively.