GitHub Copilot trades the IDE for a terminal tab

What it does

GitHub Copilot CLI is a terminal-native interface to the same agentic harness that powers GitHub Copilot. You chat in natural language to build, edit, debug, or refactor local code, and the agent previews every proposed action before executing it. It also hooks into your GitHub account to read repositories, issues, and pull requests.

The interesting bit

Unlike passive autocomplete, it runs an agentic backend that plans multi-step tasks, and it ships with GitHub’s MCP server by default while allowing custom MCP servers to extend its reach. Every prompt costs one premium request from your Copilot quota, so the meter is always running.

Key highlights

• Requires an active GitHub Copilot subscription; enterprise admins can disable access at the org level.
• Defaults to Claude Sonnet 4.5, with Claude Sonnet 4 and GPT-5 available via the /model command.
• Supports LSP for go-to-definition and diagnostics, but you must install and configure language servers yourself in ~/.copilot/lsp-config.json or per-repo .github/lsp.json.
• Includes an experimental “Autopilot” mode that nudges the agent to keep working until a task is finished.
• Actions are gated behind explicit user approval; nothing executes blindly.

Caveats

• The project is explicitly early-stage and iterating rapidly, so expect frequent changes and rough edges.
• Each prompt consumes a premium request, which could get expensive if you treat it like a casual shell assistant.
• LSP servers are not bundled, so out-of-the-box code intelligence depends on manual configuration.

Verdict

Worth a look if you already pay for Copilot and prefer living in the terminal over an IDE. Skip it if you expect a free, drop-in replacement for your shell or want fully autonomous agent execution.