Skip to content
Agent Radar

Microsoft Agent Framework 1.24.0 sets Python 3.11 baseline

Updated Microsoft Agent Framework

Summary

Agent Framework 1.24.0 for Python (Oct 30, 2026) sets a Python 3.11 baseline, breaks the beta Redis context index and adds an MCP approval default.

Tool
Microsoft Agent Framework
Category
Frameworks
Date
Open source
Yes
MCP
Yes
Pricing
Not stated in the sources we cited.
Sources

Microsoft tagged Agent Framework python-1.24.0 on October 30, 2026, eight days after 1.23.0. The agent-framework package reached PyPI the same evening (UTC). Two notes at the top of the release need attention before you upgrade: a breaking change in the beta Redis context provider, and a new Python baseline.

Release facts

FactDetail
ProductMicrosoft Agent Framework, Python packages
Version1.24.0 (GitHub tag python-1.24.0)
Release dateOctober 30, 2026 (GitHub release and PyPI upload)
Previous release1.23.0, October 22, 2026
LicenseMIT, per the GitHub API
Installpip install agent-framework==1.24.0

Before you upgrade

  • Redis context index (breaking). The beta RedisContextProvider now keeps exact, case-sensitive scope identity and rejects the old raw-scope index schema. The release notes say to delete the old context index and its hashes, let the provider create a new index, and re-ingest your data. overwrite_index=True does not migrate it (PR 9461).
  • Python 3.11 is the baseline. Python 3.10 can still install 1.24.0, but compatibility is no longer guaranteed, and the next release removes 3.10 from Requires-Python (PR 9824).
  • Minimum versions moved. Provider packages now require Core 1.24; agent-framework-claude needs Claude Agent SDK 0.2.140 or later; the Oracle package needs python-oracledb 3.0; MongoDB and DocumentDB need PyMongo 4.15.

What's new

  • Explicit MCP approval default. Local MCP tools and prompts that are not in a per-tool approval list can get an explicit default approval mode, and you get a warning when never_require_approval leaves that default implicit (PR 9828).
  • Agent-triggered compaction (experimental). AgentTriggeredCompactionStrategy and UnloadSkillsCompactionStrategy let an agent compact its context or unload skills mid-run (PR 9384).
  • Async conditions in switch-case workflow edge groups (PR 9669).
  • Owned connection cleanup. Chat clients for OpenAI, Anthropic, Bedrock, Foundry, Gemini, Mistral and Ollama gain optional async contexts and close only the connections they created (PR 9748).
  • Telemetry: a shared microsoft.agent-framework resource attribute on logs, traces and metrics (PR 9589).
  • Also: a timeout parameter on FoundryChatClient and an alpha NVIDIA NIM embedding client.

Notable fixes

  • Restricted checkpoint loading now rejects metaclasses, and Cosmos checkpoints that cannot be restored under the type allowlist are refused (PRs 9825, 8875).
  • Declarative agents honor a registered function's approval requirement even when the action omits it (PR 9606).
  • The shell tool applies output limits to standard error as well as standard output (PR 9795).
  • Citations survive replay across providers, and the Claude integration now sends declared tool JSON schemas (PRs 9436, 9302).

Questions

Should I upgrade now? Yes if you are on Python 3.11 or later and do not use RedisContextProvider. If you do use it, plan the index rebuild and re-ingest first, because the provider will reject the old index.

I still run Python 3.10. What happens? 1.24.0 installs, but Microsoft no longer supports that combination, and the next release will refuse to install. Pin agent-framework<1.25 until you move to 3.11.

I'm on Semantic Kernel. Is this my upgrade path? Microsoft says so: the Semantic Kernel README calls Agent Framework its successor and links a migration guide. See Semantic Kernel 1.46.0 for the latest release on that side, and the frameworks category for AG2 and LangGraph.

Alternatives

Microsoft Agent Framework compared with its alternatives
ToolLicensePricing note
Microsoft Agent FrameworkThis launchMicrosoft Agent Framework is an MIT-licensed Python and .NET framework for AI agents and multi-agent workflows, and the successor to Semantic Kernel.
MITMicrosoft Agent Framework is MIT-licensed and free to install from PyPI or NuGet. You pay your model provider, or Azure if you host agents on Microsoft Foundry.
Semantic KernelSemantic Kernel is Microsoft's MIT-licensed SDK for AI agents and plugins in Python, .NET and Java, now succeeded by Microsoft Agent Framework.
MITSemantic Kernel is MIT-licensed and free to install from PyPI or NuGet (Java builds live in microsoft/semantic-kernel-java). You pay your model provider for tokens.
AG2AG2 (formerly AutoGen) is an Apache-2.0 Python framework for AI agents and multi-agent networks, installed from PyPI as ag2 and imported as ag2.
Apache-2.0AG2 is Apache-2.0 and free to install from PyPI. You pay your model provider for tokens.
LangGraphLangGraph is an open-source orchestration runtime for building stateful, resilient agents.
MITNot stated

More from Microsoft Agent Framework

View all

  • Agent Framework 1.25.0 drops Python 3.10, caps MCP

    Microsoft Agent FrameworkAgent Framework 1.25.0 for Python (Nov 7, 2026) requires Python 3.11, bounds MCP discovery and timeouts, scopes workflow agents and renames workflow spans.

Questions

What is Microsoft Agent Framework?

It is Microsoft's open-source framework for building, orchestrating and deploying AI agents and multi-agent workflows. The Python package is agent-framework on PyPI and the .NET package is Microsoft.Agents.AI on NuGet. A Go SDK lives in a separate repository, microsoft/agent-framework-go.

Is Microsoft Agent Framework the same as Semantic Kernel?

No, it is the successor. The Semantic Kernel README now says Semantic Kernel is Microsoft Agent Framework and calls MAF its enterprise-ready successor. Microsoft publishes migration guides from both Semantic Kernel and AutoGen on Microsoft Learn.

How do I install Microsoft Agent Framework?

pip install agent-framework installs the standard package set for Python. For .NET, dotnet add package Microsoft.Agents.AI. Experimental modules ship separately as agent-framework-lab.

Which Python version does it need?

Python 3.11 or later. Release 1.24.0 made 3.11 the supported baseline, and 1.25.0 (November 7, 2026) removed Python 3.10 installation eligibility from every package.

Do I have to use Azure?

No. The README lists Microsoft Foundry, Azure OpenAI, OpenAI and the GitHub Copilot SDK, and the 1.24.0 release touches separate client packages for Anthropic, Bedrock, Gemini, Mistral, Ollama and NVIDIA. The Python quickstart happens to use Foundry with an Azure CLI login.

Filed on Agent Radar. Claims in this note are limited to the sources above.