.NET: Python: Azure Functions feature branch

[Dmytro Struk (dmytrostruk)]

This PR merges the feature-azure-functions branch into main. More info on this feature below:

Durable Task extension for Microsoft Agent Framework

This PR introduces two new .NET packages and one new Python package, enabling new scenarios for MAF on Azure Functions. In particular, this extension enables hosting MAF agents in distributed, serverless environments and integrates them natively into the Durable Functions orchestrations.

Documentation

• Detailed MSLearn docs and tutorials: https://github.com/MicrosoftDocs/semantic-kernel-pr/pull/755

Added NuGet Packages

• Microsoft.Agents.AI.DurableTask: Core logic for the durable extension that contains the main abstractions and can be self-hosted.
• Microsoft.Agents.AI.Hosting.AzureFunctions: Azure Functions integration, including built-in HTTP triggers, MCP tool triggers, etc.

Added Python Packages

• agent-framework-azurefunctions: Azure Functions integration as well as the core durable extension abstractions. This may be split into two packages (similar to .NET) in the future.

Added Samples

Several new samples have been added for both .NET and Python demonstrating how to use this new extension, covering several use cases. The samples also include detailed READMEs, guiding users through the process of running the samples locally.

Added Tests

Both the .NET and Python changes include unit and integration tests. The integration tests also take care of validating the new samples we added above, ensuring that they continue to work as we make changes. The new integration tests depend primarily on Azure OpenAI as the external resource, use Azure CLI credentials (no secrets), and have been stable for several days.

Test Modifications

In order to run our integration tests, we've updated dotnet-build-and-test.yml and python-merge-tests to run a new action, azure-functions-integration-setup which starts up two emulators (Azurite for Azure Storage and the Durable Task Scheduler emulator) as well as installing the Azure Functions Core Tools, which is used for running samples validation. This action only runs for integration tests and is not expected to contribute much to the length of the CI process.

Feature Branch PRs

• Python: Add Scaffolding for Durable AzureFunctions package to Agent Framework #1823
• .NET: Durable extension: initial src and unit tests #1900
• Python: Add Durable Agent Wrapper code #1913
• Azure Functions .NET samples #1939
• Python: Add Unit tests for Azurefunctions package #1976
• .NET: [Feature Branch] Migrate state schema updates and support for agents as MCP tools #1979
• Python: Add more samples for Azure Functions #1980
• .NET: [Feature Branch] Durable Task extension integration tests #2017
• Python: Add Integration tests for AzureFunctions  #2020
• .NET: [Feature Branch] Update OpenAI config for integration tests #2063
• .NET: [Feature Branch] Update dotnet-build-and-test.yml to support integration tests #2070
• [Feature Branch] GitHub actions gets stuck waiting for DTS emulator to start up #2103
• .NET: [Feature Branch] Introduce Azure OpenAI config for .NET pipeline #2106
• Python: Fix Http Schema #2112
• .NET: [Feature Branch] Fix or remove broken markdown file links #2115
• Python: Fix AzureFunctions Integration Tests #2116
• .NET: [Feature Branch] Update HTTP API to be consistent across languages #2118
• Python: Move azurefunctions to azure for import #2141
• .NET: [Feature Branch] Remove IsPackable=false in preparation for durable extension nuget release #2142
• .NET: Minor serialization updates from review feedback #2150
• Python: Update azurefunctions version and address comments from #1916 #2160

[Copilot]

Pull Request Overview

This PR adds a new Python package for Azure Functions integration with the Microsoft Agent Framework. The package enables hosting Agent Framework agents on Azure Durable Functions with features like state persistence, conversation history replay, and automatic failure recovery.

• Adds a new agent-framework-azurefunctions package under python/packages/azurefunctions/
• Updates the uv.lock file to include the new package and its dependencies (azure-functions, azure-functions-durable, furl, orderedmultidict, werkzeug)
• Updates transitive dependency versions (fsspec, google-auth, polars, polars-runtime-32, sse-starlette)

Reviewed Changes

Copilot reviewed 5 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
python/uv.lock	Adds new package dependencies for Azure Functions integration and updates several transitive dependencies to newer versions
python/packages/azurefunctions/pyproject.toml	Defines project metadata, dependencies, and build configuration for the new Azure Functions package
python/packages/azurefunctions/agent_framework_azurefunctions/init.py	Creates package entry point with copyright notice
python/packages/azurefunctions/README.md	Provides installation instructions and basic usage documentation for the Azure Functions integration
python/packages/azurefunctions/LICENSE	Adds MIT license file for the new package
.github/CODEOWNERS	Assigns code ownership for the Azure Functions package

[Mark Wallace (markwallace-microsoft)]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/azurefunctions/agent_framework_azurefunctions
_app.py	315	81	74%	165–166, 171–172, 361, 369–370, 390–392, 398–400, 406–408, 440–441, 455, 466, 483, 486, 495–497, 499–501, 503, 505–507, 518, 520–523, 525, 527–528, 530, 537–538, 540–541, 543–544, 546, 550, 560–562, 564–565, 567–569, 576, 578–579, 581, 585, 597, 667, 679, 686–688, 733, 747, 758–760, 762–765, 790, 797, 799, 802
_callbacks.py	13	0	100%
_entities.py	199	22	88%	96, 144–146, 216, 224, 229–230, 257–258, 267, 274, 285, 291–292, 311–312, 382, 415–417, 419
_errors.py	5	2	60%	12–13
_models.py	164	12	92%	202, 205, 208, 217, 220, 223, 242, 245, 250, 255, 272, 319
_orchestration.py	67	9	86%	199, 204–211
_state.py	62	2	96%	152, 169
TOTAL	14655	2123	85%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
2028	129 💤	0 ❌	0 🔥	39.365s ⏱️

[Chris Gillum (cgillum)]

I've updated PR description to remove the placeholder and add more detailed information.