GitHub, an AI developer platform, said on Thursday it has released the GitHub Copilot SDK, a solution to support building agentic workflows, as a technical preview.

The company said the GitHub Copilot SDK is designed to reduce the burden of having to build agentic workflows from scratch. It supports using the Copilot execution engine as a headless, programmable layer across all applications.

Building agentic workflows has been very complex. That is because developers had to consider inter-turn context management, orchestration of tools and commands, setting up routing between models, integrating MCP servers, configuring permissions and security boundaries, and handling failure scenarios such as outages.

The company said the GitHub Copilot SDK gives developers programmatic access to the same production-tested execution loop used to run the GitHub Copilot CLI. It said developers can embed agentic workflows into applications without separately configuring a planner, tool execution loop and runtime, allowing them to focus on the product logic they want to implement.

Developers can also use features provided by Copilot CLI, including choosing among multiple AI models, defining custom tools, integrating MCP servers, GitHub authentication and real-time streaming. The GitHub Copilot SDK currently supports Node.js, Python, Go and .NET first. Developers can use an existing GitHub Copilot subscription or use their own API key under a bring-your-own-key approach.