azure-webapp-mcp

Template Repository to use Azure Web Apps as MCP Server (Python FastMCP)

For more details, see: Azure Web App

For the Azure Web App approach of hosting a Python MCP Server, we're more flexible and we can use the officially supported FastMCP library for this:

https://github.com/jlowin/fastmcp

Prerequisites

• Python version 3.11 or higher
• uv (https://docs.astral.sh/uv/): brew install uv
• Azure CLI
  The easiest is to install the Azure CLI Tools using Homebrew:
  • Azure CLI: brew update && brew install azure-cli
  • Azure Developer CLI: brew tap azure/azd && brew install azd
• Docker desktop https://docs.docker.com/desktop/setup/install/mac-install/ or Colima (MIT License)

  brew install colima
  brew install docker  # docker is still required for colima

For Colima you also need BuildX:

brew install docker-buildx

Then to ensure Buildx plugins are available in the docker CLI, especially for Colima, you need to update ~/.docker/config.json and add this:

"cliPluginsExtraDirs": ["$HOMEBREW_PREFIX/lib/docker/cli-plugins"]

You probably need to replace $HOMEBREW_PREFIX with the output of brew --prefix

Setup locally

1. Create a repo based on our template: https://github.com/unic/azure-webapp-mcp
2. Run uv sync
3. Then:

   uv run uvicorn server:app --host 0.0.0.0 --port 8000

   to start the server locally at port 8000.

You can then run uv run client.py in a new terminal to see if it works:

CallToolResult(content=[TextContent(type='text', text='Hello, John!', annotations=None, meta=None)], structured_content={'result': 'Hello, John!'}, data='Hello, John!', is_error=False)

Local Debugging

You can run the local server in debug mode in VSCode (instead of using uv run…), set a breakpoint somewhere and then again run the client in another terminal. The execution should stop at the breakpoint.

Setup Resources in Azure

Follow these steps in order:

1. https://uniccom.atlassian.net/wiki/spaces/INNOLAB/pages/576061498
2. https://uniccom.atlassian.net/wiki/spaces/INNOLAB/pages/576028690
3. https://uniccom.atlassian.net/wiki/spaces/INNOLAB/pages/576028740

Deploy to Azure

Once all resources are created, we can deploy the MCP server to our Azure Web App:

./deploy.sh

If there is an issue with the permissions, you might need to run:

chmod +x deploy.sh

It will then ask you to login, choose a subscription and enter:

• Name of Resource Group
• Name of the Container Registry
• Name of the Web App
• Tag of your new docker image
• Name of the new docker image

Once deployed, restart the Web App in the portal.

Testing the deployed MCP-Server

Now you can test the deployed Server in different ways, for example:

Using the client.py

Change the URL to the domain of the web app, e.g.:

https://test-mcp-web-app-gzfkgbgefud0cbc3.switzerlandnorth-01.azurewebsites.net/mcp/

Using VSCode

CMD+Shift+P using "MCP: Add Server", HTTP, enter the whole url with /mcp/ at the end, give it a name, and choose workspace or local. Then in agent mode in the chat window you can test it. The entry in the config in VSCode should look like this:

"test-mcp-web-app": {
    "url": "https://test-mcp-web-app-gzfkgbgefud0cbc3.switzerlandnorth-01.azurewebsites.net/mcp/",
    "type": "http"
}

Using Claude desktop

Open the claude_desktop_config.json and add this to the "mcpServers" object:

"test-mcp-web-app": {
  "command": "npx",
  "args": [
    "mcp-remote",
    "https://test-mcp-web-app-gzfkgbgefud0cbc3.switzerlandnorth-01.azurewebsites.net/mcp/"
  ]
}

Using Copilot Studio

Following this guide: https://uniccom.atlassian.net/wiki/spaces/INNOLAB/pages/575636509/Copilot+Studio+MCP, you can add it to Copilot Studio.