Microsoft Agent Framework Trails LangChain in GitHub Stars

A production release that hasn't yet won the popularity contest

Microsoft Agent Framework (MAF) reached general availability as version 1.0 on April 3, 2026, for both .NET and Python. Microsoft describes it as a production-ready, open-source framework for building agents and multi-agent workflows. 4 The release moves the project out of release-candidate status and adds stable APIs, versioned releases and a commitment to long-term support. 4

The community numbers tell a more modest story. As of June 2026, MAF had about 11,200 GitHub stars and was shipping releases weekly. 1 LangChain, by comparison, was past 110,000 stars by mid-2025 2 and is widely described as the most adopted framework, with roughly 90 million monthly downloads. 3 Stars measure attention rather than deployment, but a tenfold gap is still a clear signal of where developer mindshare sits.

What Microsoft actually shipped

MAF is the official successor to two earlier Microsoft projects, AutoGen and Semantic Kernel. Both were built by the same teams that now run MAF, and AutoGen has moved into maintenance mode. 1 The idea is to combine the two. AutoGen contributes its multi-agent patterns, including handoff, group chat and Magentic-One. Semantic Kernel contributes its enterprise features: graph-based workflows with checkpointing, session state, middleware and OpenTelemetry observability. 1

Microsoft's GA announcement points to several headline features: 4

• Orchestration patterns: stable sequential and group-chat orchestration.
• A2A protocol: an agent-to-agent messaging protocol that lets agents on different runtimes coordinate, so a Python agent can work with one running in .NET.
• MCP support: full Model Context Protocol integration, so agents can discover and call external tools without hand-written integration code.

The accounts differ in places. One comparison lists Python, C# and Java support and dates the public preview to October 2025. 2 Microsoft's own GA materials and an independent profile mention only .NET and Python. 14 Another overview places the AutoGen–Semantic Kernel merger announcement in 2026. 3 Readers should treat the Java claim with caution until Microsoft documents it directly.

Why the star count understates and overstates at once

The gap partly reflects history. AutoGen alone collected more than 38,000 GitHub stars before the merger. 2 MAF's 11,200 reflects a fresh repository, not the total community Microsoft has built over several years of agent tooling. Consolidating two projects into a new codebase resets that public counter, even when many of the same developers carry over.

That argument has limits. Developers who starred AutoGen have not necessarily moved to MAF. Comparison guides still list AutoGen as "transitioning," a phrase that tells teams to expect migration work rather than a smooth upgrade. 3 Every consolidation risks losing users who decide that if they must rewrite anyway, they might as well evaluate competitors.

The competition is strong. LangChain leads on raw adoption. 3 CrewAI has taken a different lane: it is reportedly used by 60% of Fortune 500 companies and runs more than 450 million workflows a month. 3 That claim goes straight at the enterprise market Microsoft is targeting. The broader market is also splitting by specialty, with LlamaIndex favored for retrieval-augmented generation, CrewAI for multi-agent work and Mastra for TypeScript. 3

Where Microsoft's bet actually lies

MAF is not trying to win the way LangChain did. LangChain's appeal is vendor neutrality: it works with any model provider and any deployment target. 2 MAF's pitch rests on native integration with Azure, Microsoft 365 and Entra ID, and one comparison concludes it is the stronger choice for Azure-native organizations. 2 That audience is large, but it does not hand out GitHub stars the way independent developers and startups do.

From that angle, the star gap matters less than the GA label. Large companies tend to pick frameworks based on long-term support, stable APIs and observability, all of which version 1.0 provides. 4 Weekly releases suggest the team is moving quickly. 1 Open protocols such as A2A and MCP also reduce the lock-in concern that might otherwise push buyers toward neutral tools. 4

Still, Microsoft should not dismiss the gap. Open-source frameworks win through tutorials, integrations and community answers, and those follow developer attention. MAF has the enterprise features and the support commitment. What it lacks is the grassroots momentum that made LangChain the default. Whether Azure customers can generate that momentum on their own is the open question for the rest of 2026.