GitHub Copilot CLI vs Claude Code: how to choose in 2026

GitHub Copilot CLI vs Claude Code is a choice between a multi-vendor agent tied to your GitHub plan and a Claude-only agent with deeper controls. Copilot CLI is GitHub’s terminal agent: you sign in with GitHub, it uses your Copilot subscription and lets you pick models from several vendors, Claude included. Claude Code is Anthropic’s: you sign in with a Claude plan or an API account, it runs Claude models and adds permission modes, hooks, skills and subagents around them. Both take an effort level, resume sessions with --resume, connect to MCP servers and read CLAUDE.md.

This post compares sign-in and pricing, models, effort, permissions and sandboxing, session resume, MCP and install, then shows how to run both side by side. No benchmark scores; with Copilot you can run the same Claude model in both, so the difference is mostly the tool around it.

Copilot CLI and Claude Code at a glance

Sign-in and pricing

Copilot CLI

Copilot CLI signs in with your GitHub account through copilot login or /login in a session, and it needs an active Copilot subscription. If you are on a company plan, an admin can turn the CLI off for the organization.

At the time of writing (October 2026), GitHub lists Copilot Free ($0, limited usage), Pro ($10/month), Pro+ ($39/month), Max ($100/month), and Business and Enterprise for organizations. Usage is billed in GitHub AI Credits, and the CLI draws on them; paid personal plans can buy extra credits. Check the current details on GitHub’s Copilot plans page.

Claude Code

Claude Code needs a Claude Pro, Max, Team or Enterprise plan or a Console (API) account; the free Claude plan does not include it. It can also run through Amazon Bedrock, Google Cloud and Microsoft Foundry.

At the time of writing (October 2026), Anthropic lists Pro at $17/month billed annually or $20 monthly, Max from $100/month, and Team standard seats at $20 per seat per month billed annually or $25 monthly. Usage limits reset in 5-hour windows with a weekly cap; see Claude Code usage limits. Current prices are on Anthropic’s pricing page.

Models and effort

Copilot’s big draw is choice. --model (or /model inside a session) picks a model from Anthropic, OpenAI, Google and other vendors, and auto lets Copilot choose. --effort (also --reasoning-effort) takes low, medium, high, xhigh or max.

Claude Code runs Claude models only. --model takes aliases such as sonnet, opus and haiku or a full model name, and --effort takes low, medium, high, xhigh or max, depending on what the model supports. If you only want Claude models, Claude Code gives you the most features around them; if you want to switch vendors per task, Copilot does that in one tool.

Permissions, modes and sandbox

Copilot CLI

• Modes: --mode interactive, plan or autopilot; --plan, /plan and /autopilot switch too, and Shift+Tab cycles through them. Autopilot keeps working without stopping for each step.
• Tool rules: --allow-tool and --deny-tool allow or block specific tools, and a deny always wins. --allow-all-tools approves every tool; --allow-all (or --yolo) approves everything.
• Presets: /permissions default, assisted or allow-all change the level in a running session.
• Sandbox: /sandbox enable turns on a local sandbox that restricts file system and network access.
• Trusted directories: Copilot asks before it works in a folder you haven’t trusted.

Claude Code

• Permission modes: --permission-mode takes default (asks), acceptEdits, plan, auto, dontAsk or bypassPermissions. --dangerously-skip-permissions skips every prompt.
• Rules and hooks: allow and deny rules in settings, plus hooks that run your own scripts before or after tool calls.
• Sandbox: built-in sandboxing on macOS, Linux and WSL2; not on native Windows.

Skipping all prompts is risky in both. Keep that for an isolated folder or container; AI coding agent security covers why.

Session resume

• Copilot CLI: copilot --resume opens a picker, --resume=ID or a session name picks one, --continue takes the latest, and /resume works inside a session. -n names a session and --session-id sets its ID.
• Claude Code: claude -c continues the latest session in the folder, claude -r resumes by ID or name, --session-id sets the ID and --fork-session starts a new branch from a resumed session.

Both have /compact to shrink a long conversation.

Instructions and MCP

Here the two fit together unusually well. Copilot CLI merges custom instructions from CLAUDE.md, GEMINI.md, AGENTS.md, .github/copilot-instructions.md, .github/instructions/**/*.instructions.md and .claude/rules/**/*.md. A repo already set up for Claude Code gives Copilot its rules without extra work.

For MCP, Copilot CLI ships with a built-in GitHub MCP server, so issues, pull requests and repos are there from the start (--disable-builtin-mcps turns it off). You add servers with /mcp add or copilot mcp, they live in ~/.copilot/mcp-config.json, and Copilot also reads a repo’s .mcp.json. Claude Code manages servers with claude mcp and uses the same .mcp.json for project-scoped servers, so one file can serve both.

Install

# GitHub Copilot CLI
npm install -g @github/copilot          # Node 22+
brew install --cask copilot-cli
winget install GitHub.Copilot
curl -fsSL https://gh.io/copilot-install | bash

# Claude Code
curl -fsSL https://claude.ai/install.sh | bash
brew install --cask claude-code
winget install Anthropic.ClaudeCode

On Windows, Claude Code also installs from PowerShell with irm https://claude.ai/install.ps1 | iex.

Strengths, honestly

Copilot CLI is strongest when your team lives on GitHub and already pays for Copilot: one bill, models from several vendors, the GitHub MCP server built in, and handy commands such as /review, /delegate and /fleet. Autopilot mode and allow and deny tool rules make it easy to set how much it does on its own.

Claude Code is strongest when you want the deepest control over a Claude model: six permission modes, hooks, skills, subagents, session forking and a large ecosystem built around CLAUDE.md. Its limit is that it is Claude only and needs a Claude plan or API account.

Copilot CLI and Claude Code in AgentVera

AgentVera is a desktop app for macOS, Windows and Linux that runs 18 terminal coding agent CLIs side by side in real terminals, in a grid per project, with your existing logins. GitHub Copilot CLI and Claude Code are both in the New agent dialog.

Automatic code review is a natural fit for this pair. When a turn ends, the Claude model you choose reviews the commits from that turn or all changes, and lists findings by severity with file and line. If you turn it on, the findings go back to the agent for up to two rounds of fixes. So Copilot can write with whichever model you picked and a Claude model checks the result.

With a flow you can go further: when Copilot finishes its turn, its last reply goes into your template ({{output}}, {{source}}, {{target}}) and is sent to Claude Code, which waits its turn if busy. A turn limit stops the loop, and a Human approval node can pause for you, also on Telegram. Copilot also works in broadcasts and scheduled tasks and supports /compact.

Claude Code gets the extras that only some CLIs support: account switching with account chains that move on when the 5-hour or weekly limit runs out (Pro and Team), context profiles, automatic /compact, Remote Control and running in Docker. Account switching is not available for Copilot.

A simple setup

1. Keep your rules in CLAUDE.md (or AGENTS.md imported from it); Copilot reads both.
2. Open a Copilot CLI agent and a Claude Code agent in the same project, each in its own git worktree.
3. Register your MCP servers once in AgentVera so both get them.
4. Turn on automatic code review in project settings.
5. On Pro or Team, add a flow from one agent to the other with a review prompt and a turn limit.

The grid, worktrees, code review and the MCP manager are on the free plan. Download AgentVera and try both on your own repo. For more pairs, see Claude Code vs Codex and the best AI coding CLI tools.

Which to choose

Pick Copilot CLI if you already pay for GitHub Copilot, want to switch between models from several vendors and like having GitHub built in. Pick Claude Code if you work mostly with Claude models and want permission modes, hooks, skills and subagents. Since Copilot reads CLAUDE.md, trying both on the same repo costs you almost no setup.