Serverless MCP Orchestrator

Proof of concept for a cloud-native stack that connects AI models to South African macro data: a ASP.NET Core 10 REST API (Azure App Service, Azure SQL) plus a FastMCP Python layer on Azure Functions (streamable HTTP) for LLM tools.

Overview

This project explores connecting AI models to economic data through a layered architecture, with infrastructure defined in Bicep and deployed via the Azure Developer CLI (azd).

Architecture

Infrastructure details: infra/main.bicep (SQL Basic, Linux App Service, Consumption Functions, Storage, Log Analytics, Application Insights).

Run locally

API (SQL Server / LocalDB): set ConnectionStrings:ZarFlowDb, then dotnet run --project src/EconomicDataService.

MCP: cd src/McpOrchestrator, set API_BASE_URL to the API base URL, then python server.py (listens on http://127.0.0.1:8765/mcp).

Endpoints

Azure (deployed)

Run azd show for the current api and mcp-server HTTPS bases and an Azure Portal link to the resource group. Host names include a unique suffix per resource group.

Example bases (environment sarb-orchestrator-99):

REST API paths (/api/v1)

Legacy: GET /api/indicators (same summary as /api/v1/indicators/summary).

In Development only, OpenAPI is mapped at /openapi/v1.json (see Program.cs).

Deploy to Azure

Deployments use azure.yaml and azd.

Prerequisites

• Azure subscription with permissions to create the resources in infra/main.bicep
• Azure Developer CLI — azd auth login
• .NET 10 SDK and Python 3.11+ for local builds

First-time provision

From the repository root:

1. azd init if you do not yet have a local azd environment (choose a name you will reuse).
2. Set the Bicep SQL administrator password: azd env set adminPassword "<your-strong-password>"
3. Run azd up — provisions infrastructure, deploys the API and MCP Function app, and injects settings. EF Core migrations run on API startup against the deployed database.

Deploy application updates only

azd deploy --no-prompt

Show endpoints and portal link:

azd show

GitHub Actions

Workflow .github/workflows/azure-dev.yml runs azd deploy on push to main and workflow_dispatch. Set repository variable AZURE_ENV_NAME to your azd environment name, and secrets AZURE_CLIENT_ID, AZURE_TENANT_ID, AZURE_SUBSCRIPTION_ID for azure/login (OIDC).

CI: .github/workflows/ci.yml (dotnet test, pytest).

Security and configuration

• No secrets committed: use .gitignore, user secrets locally, and azd / Key Vault references for deployment (see infra and environment config).
• Connection strings for the API are applied via App Service settings from Bicep.

Troubleshooting

• **403 Site Disabled during API dep

... View full README on GitHub