Related Pages

Related topics: LSP Servers, MCP, and Extensibility Configuration, Authentication, Model Selection, and Daily Workflows

Overview, Prerequisites, and Installation

Purpose and Scope

GitHub Copilot CLI is a terminal-native agentic assistant that brings the same coding-agent harness used by GitHub's Copilot coding agent into a local command-line interface. It is designed for synchronous, in-repo work: you launch the binary inside a project folder, converse in natural language, and the agent plans and executes coding actions (build, edit, debug, refactor) while you preview and approve every step. Source: README.md:1-18.

The tool is positioned as an early-stage product that is iterating quickly. Per the project's own framing, it "brings AI-powered coding assistance directly to your command line, enabling you to build, debug, and understand code through natural language conversations," and it is "powered by the same agentic harness as GitHub's Copilot coding agent." Source: README.md:3-5.

Key capabilities advertised in the README include:

• Terminal-native workflow: No IDE or browser context-switching required. Source: README.md:12-13.
• GitHub integration: Repositories, issues, and pull requests are reachable via natural language using the same GitHub account you already use. Source: README.md:14-15.
• Agentic execution: The agent can plan and execute multi-step coding tasks rather than only suggesting snippets. Source: README.md:16-17.
• MCP extensibility: A built-in GitHub MCP server ships by default, with support for custom MCP servers. Source: README.md:18-19.
• Explicit approval: Nothing executes without user confirmation. Source: README.md:20-21.

Related Community Signals

Several high-engagement issues shape the roadmap around the install/launch experience:

• Model selection (#32): Users want to use 0x-credit free models (GPT-4.1, GPT-4o, GPT-5 mini, Grok Code Fast) instead of being locked to a premium default. Today, switching models is done post-launch via the /model slash command. Source: README.md:112-114.
• Legacy gh copilot workflows (#53): A long-running thread requests that the legacy gh copilot subcommands be restored so existing scripts do not break. The CLI ships as a standalone copilot binary today. Source: README.md:98-106.
• ACP (Agent Client Protocol) (#222): An open request to integrate ACP so editors can drive the same agent over a standard protocol.
• Custom slash commands (#618): Request to load user-defined prompts from .github/prompts/, mirroring the VS Code extension.

Prerequisites

Supported Platforms

Copilot CLI ships binaries for the three major desktop platforms:

Source: README.md:30-34.

Account and Subscription

An active Copilot subscription is mandatory. Organization and enterprise administrators can disable the CLI for their members; if that has happened, the CLI cannot authenticate regardless of install success. Source: README.md:36-41.

Licensing Constraint

The GitHub Copilot CLI License explicitly prohibits modifying the binary, creating derivative works, or redistributing modified copies. Redistribution is permitted only unmodified, bundled inside a larger application, and not on a standalone basis. This is relevant for anyone planning to package or fork the tool. Source: LICENSE.md:1-26.

Installation Methods

The project ships four parallel install paths so users can pick whichever matches their platform and tooling preference.

1. Install Script (macOS and Linux)

The canonical install path uses a shell pipeline hosted at https://gh.io/copilot-install. Source: README.md:44-45.

curl -fsSL https://gh.io/copilot-install | bash

Equivalent wget form is also documented. Source: README.md:46-47.

Two environment variables tune the script's behavior:

• PREFIX — install location. Defaults to /usr/local under root, otherwise $HOME/.local. Prepend with sudo to install system-wide. Source: README.md:50-53.
• VERSION — pin a specific release tag (for example, v0.0.369) instead of tracking latest. Source: README.md:55-56.

Example pinning both:

curl -fsSL https://gh.io/copilot-install | VERSION="v0.0.369" PREFIX="$HOME/custom" bash

Source: README.md:57-58.

2. Homebrew (macOS and Linux)

Stable and prerelease casks are published to the Homebrew tap. Source: README.md:61-67.

brew install copilot-cli
brew install copilot-cli@prerelease

3. WinGet (Windows)

winget install GitHub.Copilot
winget install GitHub.Copilot.Prerelease

Source: README.md:70-75.

4. npm (Cross-platform)

Distributed as the @github/copilot package; works on every supported platform including Windows. Source: README.md:78-86.

npm install -g @github/copilot
npm install -g @github/copilot@prerelease

flowchart LR
    A[Pick platform] --> B{Windows?}
    B -- Yes --> C[WinGet or npm]
    B -- No --> D{Use Homebrew?}
    D -- Yes --> E[brew install copilot-cli]
    D -- No --> F[curl/wget install.sh]
    E --> G[copilot binary in PATH]
    C --> G
    F --> G

Launching, Authentication, and First Use

Once installed, invoke the binary inside any directory containing source code. The README recommends running copilot with no arguments in a project folder. Source: README.md:90-91.

Authentication Options

Two flows are documented:

1. Interactive login: On first launch, if you are not already authenticated, the CLI prompts you to run the /login slash command and follow the on-screen OAuth-style flow. Source: README.md:95-97.
2. Personal Access Token (PAT): A fine-grained PAT with the "Copilot Requests" permission can be exported via the GH_TOKEN or GITHUB_TOKEN environment variables. GH_TOKEN takes precedence. Source: README.md:100-106.

PAT setup steps are:

1. Visit https://github.com/settings/personal-access-tokens/new.
2. Under Permissions, add the Copilot Requests scope.
3. Generate and copy the token.
4. Export it: export GH_TOKEN=….

Selecting a Model

The CLI defaults to Claude Sonnet 4.5 at launch. Other models, including Claude Sonnet 4 and GPT-5, are reachable from inside an interactive session via the /model slash command. The community's open request #32 asks for a way to default to a 0x-credit free model; today the only mechanism is choosing one manually after each launch. Source: README.md:112-114.

Experimental Mode

Two ways to opt in to in-development features:

• Launch flag: copilot --experimental.
• In-session toggle: /experimental.

The setting persists in user config, so subsequent launches inherit it. The current headline experimental feature is Autopilot mode, toggled by pressing Shift+Tab, which lets the agent keep iterating until a task is complete. Source: README.md:116-127.

Common Failure Modes

• Authentication blocked by org policy: Install completes but /login fails; check organization Copilot policy settings. Source: README.md:36-41.
• Stale binary behavior: Because the project iterates quickly, the maintainers recommend keeping the client up to date. Old builds may behave differently from the documented surface. Source: README.md:131-134.
• PATH not updated after script install: When using PREFIX=$HOME/custom or a non-root install, add $PREFIX/bin to your shell's PATH or the copilot command will be missing.
• License-driven redistribution blocks: Forks or repackagings cannot legally be redistributed modified; only unmodified copies embedded in a larger application are permitted. Source: LICENSE.md:7-15.

See Also

• Configuring LSP Servers — Code-intelligence setup is documented as a separate concern, with config files at ~/.copilot/lsp-config.json (user) and .github/lsp.json (repo). Source: README.md:17-25.
• Changelog — Release-by-release notes including the v1.0.68 model additions and slash-command alias work. Source: changelog.md:1-9.
• Official documentation: <https://docs.github.com/copilot/concepts/agents/about-copilot-cli>. Source: README.md:6.

Note: This page reflects the publicly visible README and LICENSE content at the time of writing. Internal source files beyond README.md, LICENSE.md, install.sh, package.json, and changelog.md were not retrieved for this topic, so deeper implementation details (auth flows, MCP server internals, slash-command registry) should be covered in dedicated pages.

Related Pages

Related topics: Overview, Prerequisites, and Installation, Experimental Mode, Changelog, and Community Roadmap

LSP Servers, MCP, and Extensibility Configuration

Overview

GitHub Copilot CLI is designed as an extensible agentic harness. Beyond its core conversational abilities, it exposes three primary extensibility surfaces: Language Server Protocol (LSP) servers for in-editor code intelligence, Model Context Protocol (MCP) servers for tool and resource integration, and slash commands for user-defined prompt shortcuts. Together, these mechanisms let users tailor the agent to their local toolchains, third-party services, and personal workflows without modifying the CLI binary itself.

The CLI is shipped as a thin client that authenticates with a GitHub account, then delegates reasoning to a selected model (defaulting to Claude Sonnet 4.5) while orchestrating external tools through the protocols described below. Source: README.md.

LSP Server Configuration

Purpose and Scope

LSP integration enables code-aware features such as go-to-definition, hover information, and diagnostics inside the interactive Copilot CLI session. The CLI does not bundle language servers — users must install them separately (for example, typescript-language-server via npm install -g). Source: README.md.

Configuration Locations

LSP servers can be declared at two scopes, both of which are merged at launch:

Source: README.md.

Configuration Schema

A minimal LSP entry maps a server command to one or more file extensions:

{
  "lspServers": {
    "typescript": {
      "command": "typescript-language-server",
      "args": ["--stdio"],
      "fileExtensions": {
        ".ts": "typescript",
        ".tsx": "typescript"
      }
    }
  }
}

Each key under lspServers is a logical server name. The command and args define how the server is spawned, while fileExtensions controls which files activate it. Users can inspect the merged configuration at runtime with the /lsp slash command inside an interactive session. Source: README.md.

flowchart LR
    A[Editor / CLI Session] -->|reads| B[~/.copilot/lsp-config.json]
    A -->|reads| C[.github/lsp.json]
    B --> D[Merged LSP Config]
    C --> D
    D -->|spawns| E[typescript-language-server]
    D -->|spawns F[gopls]
    D -->|spawns G[Other servers]
    A -->|/lsp command| H[Status view]

Runtime Behavior

When a file matching a configured extension is opened or modified, the CLI launches the corresponding server over stdio. Diagnostics and other responses are surfaced back into the chat context so the agent can reason about them. If you encounter a non-functional language feature, first verify the binary is on PATH and that the file extension mapping matches the project's actual file types.

MCP Server Integration

Built-in Default

The CLI ships with GitHub's own MCP server enabled by default, providing immediate access to repository, issue, and pull-request tooling. Source: README.md.

Adding Custom MCP Servers

Custom MCP servers extend the agent with additional tools, prompts, and resources. They are configured through the /mcp slash command inside an interactive session. The CLI persists entries so they survive restarts. Source: README.md.

A long-standing community request, tracked in issue #33 "Support OAuth http MCP servers" (38 comments), highlights the demand for connecting to remote, OAuth-protected MCP endpoints such as the Figma and Atlassian remote servers. As of release 1.0.68 (2026-07-01), the focused field in the /mcp config form is now marked with a ❯ chevron in addition to color, improving accessibility when stepping through the form. Source: changelog.md.

Custom Slash Commands and Workflow Extensibility

Bundled Slash Commands

The CLI exposes a growing set of built-in slash commands. Notable examples include:

• /model — switch the active model (e.g., between Claude Sonnet 4.5, Claude Sonnet 4, and GPT-5). Source: README.md.
• /login — authenticate when no GitHub session is detected. Source: README.md.
• /lsp — view configured language servers. Source: README.md.
• /mcp — manage MCP server entries. Source: README.md.
• /experimental — toggle experimental features for the current user. Source: README.md.
• /feedback — submit a confidential feedback survey. Source: README.md.

Tab Completion and Aliases

Release 1.0.68 introduces inline alias rendering during tab completion, so pressing Tab after a partial command displays alternatives such as /pr automerge|agentmerge. This is documented in: Source: changelog.md.

Custom Prompts from `.github/prompts`

Community issue #618 "Feature Request: Support custom slash commands from .github/prompts directory" (31 comments) asks for parity with the VS Code extension's behavior of loading prompt files from a .github/prompts/ directory. Users should monitor the changelog and the issue thread for status updates before relying on this mechanism. Source: README.md and changelog.md.

Experimental Mode and Stop Hooks

Enabling Experimental Features

Experimental features are gated behind a feature flag activated by either copilot --experimental at launch or the /experimental slash command. Once toggled on, the setting is persisted in the user's config so subsequent launches inherit it. The headline experimental capability is Autopilot mode, cycled via Shift+Tab, which encourages the agent to continue working until a task is complete. Source: README.md.

Stop Hooks

Release 1.0.68 adds support for stop hooks, allowing users to define actions that fire when the agent finishes a turn. Combined with Shift+Tab mode cycling, this provides finer-grained control over long-running autonomous sessions. Source: changelog.md.

Common Failure Modes

1. LSP server not found — the binary is missing from PATH. Install it globally and restart the CLI.
2. MCP server fails to connect — verify the command path and arguments via /mcp; for remote/OAuth servers, confirm tokens and scopes (see issue #33).
3. Wrong model selected — each prompt reduces your monthly premium-request quota; use /model to choose a 0x-cost option where available (see issue #32 "Ability to use Free Models").
4. IDE disconnect — during transient disconnects the CLI now keeps IDE tools available and returns a clear error, automatically recovering when the IDE reconnects. Source: changelog.md.

See Also

• README.md — Installation, authentication, and high-level feature overview.
• changelog.md — Per-release notes including MCP, LSP, and stop-hook changes.
• LICENSE.md — Terms governing redistribution of the CLI binary.
• Official documentation: About Copilot CLI.

Related Pages

Related topics: Overview, Prerequisites, and Installation, Experimental Mode, Changelog, and Community Roadmap

Authentication, Model Selection, and Daily Workflows

Overview

GitHub Copilot CLI is an agentic AI coding assistant that runs directly in the terminal. It is powered by the same agentic harness used by GitHub's Copilot coding agent, providing intelligent, GitHub-aware assistance while requiring an active Copilot subscription. This page documents the three core operational concerns for daily use: how to authenticate, how to choose a model, and how to drive the CLI through its built-in slash-command workflow. Source: README.md.

Because the CLI is in active development and ships frequent releases, users are expected to keep the client up to date and to participate in feedback through issues, Discussions, or the in-CLI /feedback command. Source: README.md.

Related Pages

Related topics: LSP Servers, MCP, and Extensibility Configuration, Authentication, Model Selection, and Daily Workflows

Experimental Mode, Changelog, and Community Roadmap

Purpose and Scope

GitHub Copilot CLI is a terminal-native AI coding agent built on the same agentic harness as GitHub's Copilot coding agent. Because the project is iterating quickly, the maintainers distinguish between stable behavior (shipping by default) and experimental behavior (opt-in, subject to change). This page consolidates three closely related concerns:

1. How the Experimental Mode is toggled and what it unlocks.
2. How the changelog communicates release-level changes (model additions, IDE resilience, slash-command UX, hook system).
3. The community roadmap — the highest-engagement feature requests that drive near-term planning.

All information below is sourced from the repository's README.md, LICENSE.md, the publicly surfaced release notes, and the open issue tracker.

Experimental Mode

Experimental Mode is an opt-in toggle that grants access to features still under active development. It is described in the introductory section of the README as a way to "enable access to new features that are still in development." Source: README.md

Activation

Two equivalent paths are documented:

• Launch flag: copilot --experimental starts the CLI directly in experimental mode.
• In-session command: /experimental flips the persisted preference while a session is running.

Once activated, the setting is persisted in the user's config, so subsequent launches no longer require the --experimental flag. Source: README.md

Feature: Autopilot

The headline experimental capability is Autopilot mode, a persistence-oriented mode in which the agent continues working until a task is completed. Users cycle through modes (and reach Autopilot) by pressing Shift+Tab. The README frames Autopilot as a mode that "encourages the agent to continue working until a task is completed." Source: README.md

Failure Modes and Caveats

Because experimental features are not API-stable, users should expect:

• Behavior changes between minor releases without prior deprecation notice.
• Removal of features that fail to graduate to stable status.
• Possible interaction effects with model selection (/model), MCP servers, and LSP servers.

The README does not enumerate every experimental feature; only Autopilot is explicitly named. Source: README.md

Release Cadence and Changelog Highlights

The project publishes a changelog.md file in the repository root, referenced from the README's "Configuring LSP Servers" section: "For more information, see the changelog." Source: README.md

The most recently surfaced release notes describe a steady cadence of UX, model, and resilience improvements. The following table summarizes themes that recur across recent releases:

The dominant narrative in the README is that the CLI is "still early in our journey" and that users should "keep your client up to date for the latest features and fixes." Source: README.md

Community Roadmap

The open issue tracker surfaces a small number of feature requests with disproportionately high engagement. These shape the near-term roadmap even when no maintainer response has been posted.

High-engagement Requests

• Free / 0x-credit models (#32, 45 comments) — Request to select from free Copilot models (e.g., GPT-4.1, GPT-4o, GPT-5 mini, Grok Code Fast) instead of being billed premium-request quota for the default Sonnet 4 model.
• Agent Client Protocol (ACP) support (#222, 25 comments) — Add ACP so that IDEs and editors can integrate the CLI as a standardized agent, analogous to LSP.
• OAuth HTTP MCP servers (#33, 38 comments) — Enable remote MCP servers that use OAuth (e.g., Figma Remote MCP, Atlassian MCP).
• Restore classic Copilot-in-the-CLI workflows (#53, 37 comments) — The most reacted issue; the community has started unofficial forks (e.g., shell-ai) to recover the old gh copilot commands.
• Custom slash commands from .github/prompts (#618, 31 comments) — Allow prompt files in .github/prompts/ to be exposed as custom slash commands, mirroring VS Code behavior.

Roadmap Signal

Several themes recur across these threads and align with the existing release notes:

flowchart LR
    A[Community Requests] --> B{Theme}
    B --> C[Model Selection<br/>#32]
    B --> D[IDE / Editor Integration<br/>#222]
    B --> E[Remote MCP Auth<br/>#33]
    B --> F[Custom Slash Commands<br/>#618]
    C --> G[/model picker]
    D --> H[ACP / LSP surface]
    E --> I[/mcp OAuth flow]
    F --> J[.github/prompts loader]

The release notes for 1.0.68 already show movement toward (G), (I) (OAuth-related MCP work), and (J)-adjacent slash-command UX improvements (inline aliases), indicating that community feedback is feeding directly into shipped behavior. Source: README.md; release notes for 1.0.68.

Feedback and Contribution Channels

The README closes with explicit guidance for participation: open issues, join Discussions, and run /feedback from inside the CLI to submit a "confidential feedback survey." Source: README.md

Licensing terms, which govern redistribution and modification of the CLI binary, are defined in LICENSE.md. The license explicitly does not grant the right to modify, adapt, translate, or create derivative works of the Software, which is relevant context for community forks such as those referenced in issue #53. Source: LICENSE.md

See Also

• README.md — Overview, installation, configuration
• changelog.md — Per-release notes
• LICENSE.md — License terms

Doramagic Pitfall Log

Found 22 structured pitfall item(s), including 9 high/blocking item(s). Top priority: Installation risk - Installation risk requires verification.

1. Installation risk: Installation risk requires verification

• Severity: high
• Finding: Project evidence flags a installation risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/1504

2. Installation risk: Installation risk requires verification

• Severity: high
• Finding: Project evidence flags a installation risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3331

3. Installation risk: Installation risk requires verification

• Severity: high
• Finding: Project evidence flags a installation risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/2891

4. Installation risk: Installation risk requires verification

• Severity: high
• Finding: Project evidence flags a installation risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3627

5. Configuration risk: Configuration risk requires verification

• Severity: high
• Finding: Project evidence flags a configuration risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3997

6. Configuration risk: Configuration risk requires verification

• Severity: high
• Finding: Project evidence flags a configuration risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3158

7. Security or permission risk: Security or permission risk requires verification

• Severity: high
• Finding: Project evidence flags a security or permission risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3948

8. Security or permission risk: Security or permission risk requires verification

• Severity: high
• Finding: Project evidence flags a security or permission risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3596

9. Security or permission risk: Security or permission risk requires verification

• Severity: high
• Finding: Project evidence flags a security or permission risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3982

10. Installation risk: Installation risk requires verification

• Severity: medium
• Finding: Project evidence flags a installation risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/4001

11. Configuration risk: Configuration risk requires verification

• Severity: medium
• Finding: Project evidence flags a configuration risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3984

12. Configuration risk: Configuration risk requires verification

• Severity: medium
• Finding: Project evidence flags a configuration risk. Review the linked source before relying on this workflow.
• User impact: May increase setup, validation, or first-run risk for the user.
• Recommended check: Reproduce the official install and quickstart path in an isolated environment.
• Evidence: community_evidence:github | https://github.com/github/copilot-cli/issues/3998