GitHub Copilot CLI
github.com/github/copilot-cli- Category
- Developer Tools
- Rank
- No. 339Tools index
- Listed in
- #12 Interface with your agents
- Pricing
- Paid
- Type
- TOOL
- Use case
- Coding
- Interfaces
- CLI
- Builder
- @github
- GitHub
- 11.3k stars
- Latest release
- v1.0.96-2
- Date
About
An official GitHub tool that brings Copilot's agentic coding assistant to the terminal, letting developers build, debug, and refactor code via natural language while staying connected to GitHub repos, issues, and pull requests. It supports MCP servers for extensibility, LSP integration for code intelligence, and requires explicit approval before executing actions.
What it does
GitHub Copilot CLI puts an AI coding agent directly in your terminal. You describe a task in plain language inside your project folder, and it reads the code, proposes edits, and can act on your behalf across your GitHub account: opening issues, commenting on pull requests, or making commits. It can call out to extra tools for context and to language servers for things like jump to definition, and it lists each step before running it so you approve or reject actions one at a time rather than handing over blanket control.
Why it's ranked here
This is about as close as it gets to an official reference implementation: GitHub builds it, ships it under its own proprietary license, and distributes it through an install script, Homebrew, WinGet, and npm rather than one narrow channel. It defaults to Claude Sonnet 4.5 but lets you switch models, and it layers in a broad extensibility surface: custom tool servers, language server hookups, and a plugin system covering skills and instructions. The changelog we read shows continuous, wide-ranging feature work rather than a stalled side project, which earns it a high spot among terminal coding agents.
What's good
The safety model is concrete, not aspirational: every planned action is shown before it runs, so you are approving specific edits and commands rather than trusting a black box. Installation is unusually well covered too, with a script, Homebrew, WinGet, and npm all pointing at the same binary, and the script itself checks a downloaded checksum file before ever unpacking anything. Model choice is not locked in either: the default is Claude Sonnet 4.5, but a single command switches to alternatives including GPT-5.
Tradeoffs
None of this is free or open: you need an active, paid Copilot subscription just to launch it, and every prompt you send draws down a monthly quota of premium requests, so heavy use has a ceiling built in. The license is not an open source one either. It lets you run and redistribute unmodified copies bundled inside something bigger, but forbids modifying the software or building anything derivative from it. Code intelligence is opt-in work too, since language servers are not bundled and each one has to be installed and pointed at manually.
How to use it well
This fits developers already living in the GitHub ecosystem who want an agent that can act across code and project management in one terminal session, not just edit files. Point it at a repository, let it work alongside your issues and pull requests, and lean on the approval prompts rather than disabling them, since that per-step check is the main safety net. Pair it with a language server for your primary language if you want real code intelligence, since none comes installed by default. It is not built for offline work or teams avoiding a subscription-gated assistant.
Technical notes+
install.sh detects OS and architecture, builds a release download URL for a prebuilt tarball (falling back to a git tag lookup for prerelease builds), authenticates the download with a token when one is set in the environment, verifies the tarball against a checksums file with sha256sum or shasum when that file is available, confirms the result is a real tarball before extracting, and installs into a prefix that defaults to /usr/local for root or $HOME/.local otherwise, offering to append that directory to whichever shell profile it detects. LICENSE.md is GitHub's own license rather than an OSI one: it permits running the software and redistributing unmodified copies bundled into something that adds real functionality beyond the software itself, but bars modification and derivative works outright. README.md documents that the CLI ships with GitHub's own MCP server by default while supporting additional custom MCP servers, and that LSP servers are not bundled and must be installed and configured separately at user or project scope. changelog.md's entries span sandboxed command execution, an argument parser rewritten onto the clap library, MCP OAuth flows, and a plugin system covering instructions, skills, and a marketplace listing, describing a command surface well beyond a thin chat wrapper.
Observed
- License
- GitHub's own proprietary license, not an OSI open source license; forbids modifying the software or creating derivative works
- Redistribution terms
- Unmodified copies may be redistributed only when bundled into an application or service that adds functionality beyond the software itself, never standalone
- Platform support
- Linux, macOS, and Windows
- Installation channels
- Install script, Homebrew, WinGet, and npm all distribute the same binary
- Interfaces
- Interactive terminal CLI with slash commands; acts as an MCP client (bundled GitHub MCP server plus custom servers) and an LSP client (servers installed separately)
- Install script integrity check
- The install script validates the downloaded tarball against a checksums file with sha256sum or shasum before extraction, when that file is available
Read from README.md, LICENSE.md, changelog.md, install.sh.
What it can do
Build, debug, and refactor code from natural language instructions
Natural language prompt → Code changes
Provide code intelligence via LSP integration
Source code → Code intelligence data
Intel on GitHub Copilot CLI
Tags
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.