Microsoft quietly shipped something useful this week. The Agent Framework now has generally available connectors for both GitHub Copilot and Claude Code, meaning you can build .NET and Python agents that hand off coding tasks to either system without writing adapter code yourself.
What actually changed
Before this release, if you wanted an agent to use GitHub Copilot or Claude Code, you had to write the integration. The connector layer was yours to build, test, and maintain. Every time one of these services updated their API, your integration broke. Microsoft’s Agent Framework now ships these as standard tool surfaces, so the same agent definition can route work to Copilot for repository-aware tasks or to Claude Code for general coding work.
Both connectors are GA. Not preview, not beta. Production ready from day one, with .NET and Python support.
What is Microsoft Agent Framework
If you are not familiar with it, the Agent Framework is Microsoft’s orchestration layer for building multi-agent systems. Think of it as a runtime for coordinating multiple AI agents, each of which can have different capabilities, models, and tool access. It handles the messy parts: agent discovery, task routing, state management, and error handling across agents.
The framework supports .NET and Python, which covers most enterprise development scenarios. Agents can be defined declaratively, and the runtime manages the execution flow. The new connectors slot into this model as standard tool surfaces that any agent in the system can use.
Why this matters for developers
The practical implication is simpler agent architectures. Instead of a monolithic agent that tries to do everything, you can have specialized agents — one that knows your repo structure through Copilot, another that handles open-ended coding through Claude Code, and a coordinator agent that routes work between them. The Agent Framework handles the plumbing, including retries, timeout handling, and error propagation.
This also means you can mix environments without extra work. A .NET agent can use the Copilot connector for C# work in your Azure repo, while a Python agent uses the Claude Code connector for data processing scripts. Both share the same orchestration layer, so you get consistent logging, monitoring, and error handling across the board.
Comparing the two connectors
GitHub Copilot’s connector is stronger for tasks that need repository context. It understands your codebase structure, existing code patterns, and can suggest changes that fit the project’s conventions. This makes it the right choice for refactoring, bug fixes, and incremental changes where consistency with the existing code matters.
The Claude Code connector is better for open-ended problems where you need a model to reason through a complex task from scratch, without being constrained by existing code patterns. It handles ambiguous requirements better and produces more complete solutions for greenfield work.
In practice, you would route different work to different connectors. A bug fix in an existing function? Copilot, because it knows the surrounding code. A new feature from a spec document? Claude Code, because it handles interpretation better. A multi-step task that involves both analysis and code generation? Route the analysis to Claude Code and the implementation to Copilot.
Setting up the connectors
The setup is straightforward. You define a connector in your agent configuration by specifying the type (Copilot or Claude Code) and the authentication method. The Agent Framework handles the rest. For Copilot, the connector uses your existing GitHub identity. For Claude Code, it uses the same API key configuration you would use for direct access.
Both connectors support the full capabilities of their respective services. The Copilot connector can access repository context, suggest completions, and run code reviews. The Claude Code connector can write, edit, and explain code across multiple files.
Practical considerations
Both connectors come with the usual caveats. You are sending code to an external service, so data residency and compliance rules apply. The Agent Framework handles authentication through managed identities, but you should verify that your deployment meets your organization’s data handling requirements.
Cost is another factor. Both Copilot and Claude Code are metered services on top of the Agent Framework itself. The connector eliminates integration work, but it does not eliminate the per-use cost of each service. You should budget for the combined cost of the framework and the connectors.
Start with the Agent Framework GitHub repo. The connectors are documented there with sample code for both .NET and Python. A good first test is routing a simple refactoring task through Copilot and a code generation task through Claude Code, then comparing the results. This will give you a feel for when each connector is the right choice.
What this signals for the ecosystem
Microsoft shipping a GA connector for Claude Code — an Anthropic product — is worth noting. The move suggests a pragmatic approach to the AI tooling ecosystem rather than a walled-garden strategy. Microsoft could have built an Agent Framework that only works with Copilot. Instead, they made it work with both, and the architectural choice to use standard tool surfaces means any coding agent can be plugged in as long as it exposes a compatible interface.
This is the right call. Teams do not want to be locked into one AI coding assistant, and the market is moving too fast for anyone to predict which tools will dominate in two years. An agent framework that supports multiple coding backends lets teams hedge their bets and switch as the landscape shifts.