github/copilot-cli
Developer ToolsA terminal AI coding agent that plans, edits, and debugs code with GitHub context and approval-gated actions.
Overview
GitHub Copilot CLI puts the Copilot coding agent in the terminal, letting developers build, debug, and understand code through natural-language conversations. It runs the same agentic harness as GitHub's hosted Copilot coding agent and connects to repositories, issues, and pull requests through an existing GitHub account, so work can stay local and synchronous without switching context about Copilot CLI.
The agent can plan and execute multi-step coding tasks, previews every action before execution, ships with GitHub's MCP server, and supports custom MCP servers for extension. It is aimed at developers who already use GitHub and want an agent that understands both local code and GitHub context.
Quick Start
Prerequisites: an active Copilot subscription, and PowerShell v6+ on Windows prerequisites. Organization or enterprise accounts may be blocked if an administrator has disabled Copilot CLI policy note.
On macOS or Linux, install with the script:
curl -fsSL https://gh.io/copilot-install | bash
Use | sudo bash to install to /usr/local/bin, or set PREFIX and VERSION to choose the install directory and release install script options. Alternatives are brew install copilot-cli on macOS/Linux Homebrew, winget install GitHub.Copilot on Windows WinGet, and npm install -g @github/copilot on any platform npm package.
Then launch copilot from a folder containing the code you want to work on launching. On first run, complete authentication: if you are not logged in to GitHub, use /login, or supply a fine-grained personal access token with the "Copilot Requests" permission via GH_TOKEN or GITHUB_TOKEN PAT authentication. By default the agent uses Claude Sonnet 4.5; /model switches to other available models using the CLI.
Configuration and Extras
LSP servers are not bundled. Install one separately, such as npm install -g typescript-language-server, then configure it in ~/.copilot/lsp-config.json for all projects or .github/lsp.json for one repository; /lsp shows the status LSP configuration. The same repository also ships a container image at ghcr.io/github/copilot-cli for connecting SDK clients to a headless server, not for interactive terminal use container image guide. Experimental features, including Autopilot mode, are enabled with copilot --experimental or the /experimental command and persist in the config afterward experimental mode.
Usage Constraints
Each prompt submitted consumes one premium request from the monthly quota premium requests. Interactive use expects a terminal and a code directory; the container image is a separate mode for SDK clients. The repository license is not declared as a standard recognized license in its metadata. The agent asks for approval before executing actions, but destructive commands still require user judgment.
HotRepo's Take
In the 2026-11-03 Python daily ranking, the project recorded 11 new stars and a best position of 14. That is a single snapshot, not enough to establish a trend. On HN, the public-preview thread drew 24 points and 4 comments across 2025-09-25 and 2025-09-26; one commenter, versteegen, noted that at the time /help, copilot --help, and the interface gave no way to see or switch models HN comment, while the current README documents /model and a default of Claude Sonnet 4.5. Another commenter asked how the tool handles destructive suggested commands like rm -rf HN comment, which remains a fair operational question given the approval-based design.
Written by AI from the project's README, updated . The project's own docs are the source of truth.