Tech News, Tech & Coding · Free Online Tools

IT Question Answer
Tools
Back to Web & Development
GitHub's Copilot SDK reaches general availability for building custom AI agents

GitHub's Copilot SDK reaches general availability for building custom AI agents

Web & DevelopmentBy ITQA TeamGitHub Changelog; GitHub Blog
githubcopilotsdkdeveloper toolsai agents

What GitHub announced

On June 2, 2026, GitHub announced that the Copilot SDK is generally available after a public preview that began in January. The SDK lets developers embed the same agent engine that powers Copilot into their own applications, internal tools and services, instead of building an agent loop from scratch. The release landed alongside Microsoft's Build conference. See our coverage of Build 2026 for the wider agent push.

Languages and install

Official SDKs cover Node.js and TypeScript, Python, Go and .NET, and Rust and Java were added at general availability. Packages are published through the usual channels, for example npm install @github/copilot-sdk for TypeScript and pip install github-copilot-sdk for Python.

Main capabilities

  • Model Context Protocol: connect to MCP servers, register custom tools, or override built-in tools such as search and file editing.
  • Tracing: OpenTelemetry support with W3C trace context across CLI startup, JSON-RPC calls, sessions and tool execution, which helps with debugging agent behavior in production.
  • Authentication: GitHub OAuth, GitHub Apps and environment tokens.
  • Bring your own key: use providers such as OpenAI, Microsoft Foundry and Anthropic with your own credentials.
  • Cloud and remote sessions: sessions can be backed by the cloud with repository metadata, so several clients can work with the same session.
  • Hooks: run code before or after tool use, at session start, on MCP tool calls and on permission requests.

Who can use it

The SDK is open to existing Copilot subscribers, including people on Copilot Free for personal use, and to developers without a Copilot plan who use their own model keys. GitHub did not publish separate pricing for the SDK in its announcement, so check how usage is billed for your plan before you ship anything.

Why it matters

Many teams have been wiring together their own agent runtimes with ad hoc prompt loops. A supported SDK with tracing, permissions and hooks gives them a stable base, and a stable API means fewer breaking changes than during the preview. The trade-off is dependence on GitHub's agent runtime, so teams should keep their tool definitions portable through MCP.

Advertisement