Microsoft Agent Framework Adds Stable GitHub Copilot Agent Support for .NET and Python

Copilot coding agents gain an Agent Framework integration

Microsoft has released the GitHub Copilot Agent as a stable integration for .NET and Python within Microsoft Agent Framework. The release gives developers a supported way to use GitHub Copilot’s coding-agent capabilities in applications built on the framework.

In its announcement, Microsoft said the integration is aimed at agents that need to work with codebases and development environments. These agents can read and write files, execute shell commands, fetch URLs and interact with development tools while working across repositories.

The GitHub Copilot Agent is backed by the GitHub Copilot CLI and SDK, according to Microsoft. Copilot is responsible for the central agent loop, including model calls, planning, tool invocation and session state. Microsoft Agent Framework provides the surrounding application layer, with interfaces for instructions, tools, response streaming, middleware, observability and human approval.

That split lets teams use Copilot’s repository-oriented coding runtime without implementing the execution loop themselves. It also gives developers framework-level controls that may matter in organisational deployments, such as instrumentation, custom middleware and approval steps for consequential actions.

Copilot as an Agent Framework provider

GitHub’s documentation describes the Copilot SDK integration as a provider for Microsoft Agent Framework. Developers can install and configure it as part of applications that use the framework’s agent abstractions, tool support, streaming and orchestration features.

The provider model could be useful when code work is only one part of a broader agent application. A development workflow, for example, could use Copilot to inspect a repository, edit files and run commands, while other components coordinate specialised tasks or apply approval policies before changes are accepted.

Microsoft says the framework maintains a consistent development surface around Copilot, despite Copilot retaining ownership of task planning and execution. This means teams can apply Agent Framework features around an existing coding runtime rather than replacing it with a separately built agent loop.

A focus on the software around the model

The release also reflects GitHub’s emphasis on the agentic harness: the software that supplies a model with tools, context and execution rules. In a separate GitHub Blog evaluation, GitHub reported that its shared Copilot harness delivered task-completion results comparable to model-provider harnesses in its tests while using fewer tokens in most tested configurations.

Those results are GitHub’s own measurements and should not be treated as an independent benchmark. However, they underline a growing industry view that coding-agent performance depends not only on the underlying model, but also on the systems that manage context, tools and execution.

For Microsoft Agent Framework users, the stable .NET and Python integration establishes a direct route to embed GitHub Copilot’s coding-agent runtime in applications that require additional operational visibility, extensibility and human oversight.