Microsoft Agent Framework (MAF) has shipped v1.0 support for agents backed by the GitHub Copilot SDK, moving this integration from preview to fully stable and supported across both C# and Python.

What shipped

The GitHub Copilot SDK integration is now production-ready in both MAF C# and Python SDKs. Agents built with MAF can now use the Copilot SDK as their backend, granting access to a set of coding-oriented capabilities: shell command execution, file operations, URL fetching, and Model Context Protocol (MCP) server integration. These capabilities can be woven directly into your existing development pipeline.

The v1.0 designation means the API is stable; no experimental gates or churn between preview and release. Teams can depend on this surface for production workloads.

Why this matters

Until now, building a production agent that could actually do things—run shell commands, edit files, call external tools—meant either hand-wiring each capability or choosing between fragmented frameworks. The Copilot SDK integration centralizes those primitives under MAF's orchestration layer.

The catch is scope: this is built for coding use cases first. The capabilities shipped are the ones Copilot exposes—shell, filesystem, HTTP, MCP. If your agent needs to orchestrate non-coding tasks or call domain-specific APIs, you'll still layer those yourself. But if your agents are engineering-focused—code review, refactoring, testing, deployment—the Copilot SDK backend removes a lot of plumbing.

MCP integration is the linchpin here. Instead of baking every tool into your agent at deployment time, you point it at an MCP server and it discovers tools on demand. That means a central team can publish domain expertise once, and every agent in your org picks it up without redeploy.

The multi-agent story

The same announcement also covers a separate release: "Building agent teams with Agent Framework, GitHub Copilot CLI and Squad." The pieces are landing in sequence—orchestration patterns (sequential, concurrent, group chat, handoff, magentic) all reached 1.0 across both SDKs in early July. Agent Skills for both .NET and Python are now stable. The harness (loop, planning, memory, context, approvals, telemetry) shipped on July 22. And this Copilot SDK integration completes the pipeline for teams that need agents to coordinate and act.

If you're building multi-agent systems that coordinate on coding tasks, the full stack is now stable: declare workflows in YAML, wire agents to Copilot for execution, and pull in domain expertise from MCP servers.

What to watch

The samples exist in both .NET and Python repos. The integration is stable, so start from there if you're evaluating it. The main limitation is that the Copilot SDK scope is fixed—if you need non-coding capabilities, you'll need to layer those via custom tools or other runtimes. Also watch how MCP server adoption moves; the value of dynamic skill discovery depends on a healthy ecosystem of published MCP servers.

Sources: