Create an MCP for Azure DevOps To Use With Genie Code

[emma_s]

Overview

Prompted by a customer question, I wanted to see what was possible in terms of MCP integration into Genie Code, in order to try this out I decided to look at Azure Dev Ops, as it's a common workflow to want to see your tickets alongside the work you're doing. When I dug into it I realised the Azure out of the box MCP did not have the right protocols to enable Genie code to use it. Given the common nature of this kind of request I thought I'd share my learnings here. After some testing I found the best way to achieve this was using a python udf in devops and then setting this up as a Genie Code MCP server.  If you want to give this a go, here are the instructions and code.

 What You'll Build

• Two Python UDFs in Unity Catalog that connect to Azure DevOps — search_work_items and list_project_backlog. No extra infrastructure. Add them to Genie Code via the UC Function browser and start asking natural language questions immediately.

Prerequisites

• A Databricks workspace with Apps enabled
• Databricks CLI installed and authenticated
• An Azure DevOps organisation with a project
• A Personal Access Token (PAT) for Azure DevOps with Work Items (Read) scope

Step 1: Create an Azure DevOps Project & Personal token

Create DevOps Project

1. Go to https://dev.azure.com and sign in (or create a free account)
2. Create a new organisation 
3. Create a project
4. Add some work items (Epics, Issues, Tasks) to have data to query

Create a Personal Access Token

1. Click the user settings icon (top right) → Personal access tokens
2. Click New Token
3. Name: <set-your-name>
4. Scopes: Full access (tighten for production)
5. Copy the token — you'll need it in Step 3

Step 2: Enable Prerequisites

Before creating the UC Functions, enable outbound networking for UDFs:

In your Databricks workspace, go to Settings > Preview features

Enable "Enable networking for UDFs in Serverless SQL Warehouses"

Restart your SQL warehouse after enabling. Changes can take up to 15 minutes to propagate.

Step 3: Create the UC Functions

First, create a schema to hold the functions:

CREATE SCHEMA IF NOT EXISTS <your-catalog>.<your-schema>

COMMENT 'Azure DevOps MCP tools for Genie Code';

Note: Python UDFs do not support default parameter values — all parameters are required.

Function 1: search_work_items

CREATE OR REPLACE FUNCTION <your-catalog>.<your-schema>.search_work_items(

  query STRING COMMENT 'Text to search for in work item titles and descriptions',

  max_results INT COMMENT 'Maximum number of results to return, e.g. 20'

)

RETURNS STRING

LANGUAGE PYTHON

COMMENT 'Search Azure DevOps work items by text in title or description. Use this to find tasks, issues, and epics matching a keyword.'

AS $$

import urllib.request, json, base64

PAT = "<your-pat-token>"

ORG = "<your-org-name>"

PROJECT = "<your-project-name>"

BASE = f"https://dev.azure.com/{ORG}"

creds = base64.b64encode(f":{PAT}".encode()).decode()

def api_call(path, body=None, extra_params=''):

    url = f"{BASE}/{path}?api-version=7.0{extra_params}"

    headers = {"Authorization": f"Basic {creds}", "Content-Type": "application/json"}

    if body:

        req = urllib.request.Request(url, data=json.dumps(body).encode(), headers=headers)

    else:

        req = urllib.request.Request(url, headers=headers)

    return json.loads(urllib.request.urlopen(req, timeout=30).read())

wiql = f"SELECT [System.Id] FROM workitems WHERE [System.Title] CONTAINS '{query}' OR [System.Description] CONTAINS '{query}' ORDER BY [System.ChangedDate] DESC"

result = api_call(f"{PROJECT}/_apis/wit/wiql", {"query": wiql}, f"&$top={max_results}")

ids = [wi["id"] for wi in result.get("workItems", [])]

if not ids:

    return f"No work items found matching '{query}'."

fields = "System.Id,System.Title,System.State,System.AssignedTo,System.WorkItemType"

id_str = ",".join(str(x) for x in ids[:200])

data = api_call(f"{PROJECT}/_apis/wit/workitems", extra_params=f"&ids={id_str}&$fields={fields}")

lines = [f"Search results for '{query}' ({len(data.get('value', []))} items):", ""]

for item in data.get("value", []):

    f = item.get("fields", {})

    assigned = f.get("System.AssignedTo", {})

    if isinstance(assigned, dict):

        assigned = assigned.get("displayName", "Unassigned")

    lines.append(f"- #{item['id']} [{f.get('System.WorkItemType', '?')}] {f.get('System.Title', 'Untitled')} - {f.get('System.State', '?')} (Assigned: {assigned})")

return "\n".join(lines)

$$;

Function 2: list_project_backlog

CREATE OR REPLACE FUNCTION <your-catalog>.<your-schema>.list_project_backlog(
  max_results INT COMMENT 'Maximum number of backlog items to return, e.g. 50'
)
RETURNS STRING
LANGUAGE PYTHON
COMMENT 'List the active project backlog showing all open Epics, Issues, and Tasks grouped by type.'
AS $$
import urllib.request, json, base64
PAT = "<your-pat-token>"
ORG = "<your-org-name>"
PROJECT = "<your-project-name>"
BASE = f"https://dev.azure.com/{ORG}"
creds = base64.b64encode(f":{PAT}".encode()).decode()
def api_call(path, body=None, extra_params=''):
    url = f"{BASE}/{path}?api-version=7.0{extra_params}"
    headers = {"Authorization": f"Basic {creds}", "Content-Type": "application/json"}
    if body:
        req = urllib.request.Request(url, data=json.dumps(body).encode(), headers=headers)
    else:
        req = urllib.request.Request(url, headers=headers)
    return json.loads(urllib.request.urlopen(req, timeout=30).read())
wiql = "SELECT [System.Id] FROM workitems WHERE [System.WorkItemType] IN ('Epic', 'Issue', 'Task') AND [System.State] NOT IN ('Closed', 'Removed', 'Done') ORDER BY [Microsoft.VSTS.Common.BacklogPriority] ASC"
result = api_call(f"{PROJECT}/_apis/wit/wiql", {"query": wiql}, f"&$top={max_results}")
ids = [wi["id"] for wi in result.get("workItems", [])]
if not ids:
    return "The backlog is empty."
fields = "System.Id,System.Title,System.State,System.AssignedTo,System.WorkItemType"
id_str = ",".join(str(x) for x in ids[:200])
data = api_call(f"{PROJECT}/_apis/wit/workitems", extra_params=f"&ids={id_str}&$fields={fields}")
by_type = {}
for item in data.get("value", []):
    wtype = item.get("fields", {}).get("System.WorkItemType", "Other")
    by_type.setdefault(wtype, []).append(item)
type_order = ["Epic", "Issue", "Task"]
lines = [f"Active Backlog ({len(data.get('value', []))} items):", ""]
for wtype in type_order:
    items = by_type.get(wtype, [])
    if items:
        lines.append(f"{wtype}s ({len(items)}):")
        for item in items:
            f = item.get("fields", {})
            assigned = f.get("System.AssignedTo", {})
            if isinstance(assigned, dict):
                assigned = assigned.get("displayName", "Unassigned")
            lines.append(f"  - #{item['id']} {f.get('System.Title', 'Untitled')} - {f.get('System.State', '?')} (Assigned: {assigned})")
        lines.append("")
return "\n".join(lines)
$$;

Step 4: Test the Functions

SELECT <your-catalog>.<your-schema>.search_work_items('inspection', 5);

SELECT <your-catalog>.<your-schema>.list_project_backlog(50);

Step 5: Add to Genie Code

1. Open Genie Code in your workspace
2. Switch to Agent Mode
3. Click Settings (gear icon)
4. Under "Unity Catalog Function", click Browse
5. Navigate to mcga_catalog.devops_tools
6. Select the functions and Save

Result

You can now ask Genie Code natural language questions like:

• "What's on the backlog?"
• "Find work items related to vessel inspections"
• "Search for anything about compliance"

Key Notes for UC Functions Approach

• Python UDFs cannot have default parameter values — all params must be required
• The "Enable networking for UDFs" preview must be enabled for outbound HTTP calls
• Functions use urllib.request (standard library) — no pip dependencies needed
• The Azure DevOps PAT is embedded in the function body. For production, consider using UC service credentials instead.