Copilot SDK is Now Generally Available
General availability means one thing: you can now embed GitHub Copilot's agentic engine directly into your application. The Copilot SDK surfaces the full orchestration model — planning, file editing, multi-turn sessions — as language-native libraries. You no longer need to choose between building your own agent or waiting for GitHub's UI. You can ship your own.
What the SDK exposes
The Copilot SDK surfaces agent orchestration as language-native libraries. You can create sessions, invoke tools, handle MCP protocol exchanges, and manage permission requests — without building orchestration logic from scratch. The SDK includes hooks for executing custom code before and after tool calls, allowing you to integrate custom logic into the agent's workflow.
Language coverage and ecosystem
- Node.js: First-class support for JavaScript and TypeScript applications. Use the SDK in an Express server, a Next.js app, or a CLI tool.
- Python: Synchronous and async APIs. Integrates with FastAPI, Django, and custom scripts.
- Go: Idiomatic Go interfaces for long-running agents and concurrent orchestration.
- Rust: For systems where performance and safety are non-negotiable.
- .NET: C# and F# support for enterprise Windows and cloud workloads.
- Java: Android and Spring Boot applications can embed agents directly.
Key capabilities in the SDK
Custom tools: Register your own tool definitions. The agent can call them, handle their responses, and retry on failure. Tools are discovered dynamically, so new integrations do not require SDK updates.
Model Context Protocol (MCP) integration: Connect MCP servers (Anthropic's standard for tool standardization) to the agent. This is how you wire in Stripe APIs, internal databases, or specialized services without writing binding code yourself.
Fine-grained system prompt customization: Copilot's "personality" is defined through modular prompt sections. You can override tone, safety rules, formatting preferences, and domain-specific constraints without touching the core agent logic.
Cloud and remote sessions: Sessions can live in the cloud, allowing you to start a long-running agent and query its status or results from anywhere. Sessions are addressable by URL, making it easy to share progress with team members.
Hook system: Intercept agent behavior at multiple points: before planning, after tool selection, before file edits, on completion. Use hooks to implement approval gates, audit logging, or custom resource limits.
What this changes about product development
Until now, product teams had two choices: build agent infrastructure from scratch (months of work, high complexity) or defer to GitHub's UI and accept its constraints. The SDK collapses that binary. You can now ship a custom agentic interface in your product within weeks, not months, because the orchestration layer is already written and battle-tested.
For Foculoom, this is a key unlock. We are now evaluating whether to embed Copilot SDK directly into our product interfaces rather than always routing through GitHub's UIs. That decision was impossible six months ago.
Qualifying sources
Primary source: GitHub Changelog, Copilot SDK is now generally available (published June 2, 2026).
Pricing, rate limits, and feature parity with hosted Copilot may vary. Consult the official GitHub Copilot SDK documentation before production deployment.