Why migrate from Claude Code to GitHub Copilot CLI at all?

If you're at Microsoft Experiences + Devices (Windows, M365, Outlook, Teams, Surface), Microsoft is winding down most internal Claude Code licenses by June 30, 2026 and directing engineers to GitHub Copilot CLI. If you're not at Microsoft, you usually shouldn't migrate — Claude Code with Claude Fable 5 still leads SWE-Bench Pro at 80.3% and is the better tool for hard agentic work. Cost is the only common reason for non-Microsoft enterprises to migrate.

What's the fastest path to install Copilot CLI?

Run `gh extension install github/gh-copilot` if you have the GitHub CLI installed, or download the standalone Copilot CLI binary from cli.github.com/copilot. Authenticate with `gh auth login` then `gh copilot config`. Verify with `gh copilot suggest 'list all md files modified in the last 24h'`. Total setup: ~5 minutes if you already have gh; ~15 if you need a fresh install.

What capabilities do I lose vs Claude Code?

Three main gaps as of June 2026: (1) Sub-agent maturity — Claude Code has deeper sub-agent orchestration; Copilot CLI is catching up but lags. (2) SWE-Bench Pro performance — GPT-5.5 (Copilot's default) scores ~58.6% vs Claude Fable 5's 80.3%, so hard refactors fail more often. (3) Cross-IDE breadth — Claude Code has tighter integrations with Cursor and JetBrains via shared SDK. For most day-to-day code review and chat-style help, Copilot CLI is now comparable.

Can I still use Claude Fable 5 inside Microsoft's allowed stack?

Yes, via Azure. Microsoft Foundry serves Claude Fable 5 to paying Azure customers — that's a managed access path your company likely already has. The Microsoft policy change is about ending internal Claude Code licenses (the CLI tool); it doesn't restrict you from calling Fable 5 via Azure OpenAI / Foundry endpoints in your applications. Check with your security/procurement team — internal policies may still require approval for any non-default model.

Quick Answer

How to Migrate from Claude Code to Copilot CLI (June 2026)

Published: June 13, 2026

How to Migrate from Claude Code to GitHub Copilot CLI (June 2026)

Microsoft is ending most internal Claude Code licenses by June 30, 2026. If you’re an Experiences + Devices engineer, you need to migrate your daily workflow to GitHub Copilot CLI by then. Here’s the practical migration plan: install, auth, port your agentic workflows, and what to expect when capabilities don’t translate one-to-one.

Last verified: June 13, 2026

TL;DR

Deadline: June 30, 2026 for Microsoft Experiences + Devices.
Target tool: GitHub Copilot CLI.
Time required: 30–60 minutes for individual migration; 1–2 days for team-wide ports.
Biggest gap: Hard agentic refactors (SWE-Bench Pro 80.3% → ~58.6% drop).
Workaround for Fable 5 access: Azure / Microsoft Foundry (managed Anthropic models).
Should non-Microsoft enterprises migrate? Usually no. See GitHub Copilot CLI vs Claude Code: enterprise pick.

Step 1: Install GitHub Copilot CLI (5–15 min)

If you already have the GitHub CLI:

gh extension install github/gh-copilot
gh extension update

If you don’t:

# macOS
brew install gh
brew install --cask copilot  # if Microsoft IT hasn't pre-deployed

# Windows
winget install GitHub.cli
winget install GitHub.copilot

Verify:

gh --version
gh copilot --version

Step 2: Authenticate (5 min)

gh auth login
# Choose: GitHub.com → HTTPS → Yes (Git credentials) → Login with browser
gh copilot config

For Microsoft engineers: use your @microsoft.com SSO. Copilot CLI authenticates against your Microsoft 365 Copilot license, which is already provisioned.

Verify:

gh copilot suggest "list all markdown files modified in the last 24 hours"

You should get a shell command suggestion. If you get an auth error, run gh auth status and re-login.

Step 3: Map your Claude Code workflows (15–30 min)

This is where the migration takes real thought. For every Claude Code workflow you depend on, find the Copilot equivalent.

Claude Code workflow	Copilot CLI equivalent
`claude` (start interactive agent session)	`gh copilot suggest` / `gh copilot explain` for inline; `gh copilot agent` for multi-step
`/agents` (sub-agents)	`gh copilot agent --parallel` (limited; check version)
`/hooks` (pre/post-execution callbacks)	Not yet a 1:1 match; use GitHub Actions for CI-side hooks
`/skills` (reusable procedures)	Custom slash commands via `.github/copilot-commands/`
Multi-file refactor (Composer-style)	`gh copilot agent "refactor X"` with manual scope hints
PR review automation	Copilot Workspace (native, deeper than Claude Code’s GitHub Actions)
Codebase Q&A	`gh copilot explain` + repo context
Long autonomous runs (hours)	Less mature — split into shorter tasks

The honest gaps:

Sub-agents are not as mature in Copilot CLI as in Claude Code. If you depend heavily on sub-agent orchestration, plan to split work into sequential tasks for now.
Hard refactors will retry more often. Plan for ~1.4x more iterations on tasks that previously succeeded first-shot with Fable 5.
Long autonomous runs (multi-hour) are less reliable today. Use shorter checkpointed runs.

Step 4: Configure for your team conventions (15 min)

Create a .github/copilot-commands/ directory in your repo for shared slash commands:

mkdir -p .github/copilot-commands
cat > .github/copilot-commands/review-pr.md <<'EOF'
---
name: review-pr
description: Run our team's standard PR review checklist
---
Read the diff in this PR. Check for:
1. Test coverage on new code
2. Type signatures on public APIs
3. No console.log / dbg() left in
4. Comments on non-obvious logic
5. Backward compat for any API change
Output: a markdown review with line-anchored comments.
EOF

Commit these to the repo so the whole team gets consistent agent behavior.

Step 5: Use Azure for Claude Fable 5 access (when you need it)

Microsoft’s policy ends internal Claude Code licenses, not access to Claude models. For workloads that materially benefit from Fable 5 (hard refactors, security audits, complex multi-file work), you can still call Fable 5 via:

Microsoft Foundry — managed Claude Fable 5 endpoint with Azure billing.
Azure AI Studio — Anthropic models through Azure.

This is procurement-clean: it stays inside the Microsoft / Azure billing umbrella that your org already pays. The tradeoff is that you’re not using the Claude Code CLI tool — you’re calling the model API directly from scripts or via Copilot CLI’s model-override option (when available).

Check internal policy: many teams still require explicit approval for non-default models, even via Foundry.

Step 6: Tune expectations (the honest part)

Expectation	Reality
”Copilot CLI is just as fast as Claude Code”	Inline: yes. Long agentic: no.
”I’ll never feel the SWE-Bench gap”	You will, on hard refactors. Plan retries.
”Sub-agents work the same”	Not yet. Use sequential tasks.
”PR review is worse”	Actually better — Copilot Workspace is native and tighter than Claude Code Actions.
”Cost will go up”	No — that’s the whole reason for the change.
”I’ll lose terminal UX”	Different, not worse. Copilot CLI is a clean terminal tool.

Step 7: Sunset Claude Code (June 28–30)

Before June 30:

Export any saved Claude Code sessions you want to preserve (transcript copies).
Remove claude from your shell config (.zshrc, .bashrc).
Run claude logout to clear credentials.
Uninstall: npm uninstall -g @anthropic-ai/claude-code (or your install path).
Confirm no remaining shell aliases point to claude.
Update team docs / runbooks that referenced Claude Code commands.

Common migration mistakes

Trying to one-shot the migration. Spread it across 2 weeks. Migrate one workflow, get comfortable, migrate the next.
Ignoring Copilot Workspace. It’s the closest thing Microsoft has to Claude Code’s full agentic vision. Use it for PR-side work.
Not updating team conventions. The slash command file format and shared agent prompts are different.
Assuming all Claude integrations break. Your VS Code Anthropic extension still works for non-Claude-Code use cases.

What if you’re not at Microsoft?

Don’t migrate. The Microsoft policy is Microsoft-specific:

Microsoft owns GitHub, so internal Copilot dogfooding has unique strategic value.
Microsoft’s seat-license economics dwarf any other enterprise.
Most other enterprises should run a hybrid: Copilot for breadth, Claude Code for senior-engineer depth.

See GitHub Copilot CLI vs Claude Code: enterprise pick for the full hybrid argument.

Bottom line

If you’re at Microsoft Experiences + Devices, the migration is mandatory and the practical timeline is two weekends. Install + auth + map workflows + configure shared commands + sunset Claude Code. Plan for the SWE-Bench gap on hard refactors; use Azure / Microsoft Foundry for Fable 5 access when a specific task really needs it. If you’re not at Microsoft, this migration probably isn’t for you — and Claude Code stays the better tool for the hardest parts of your job.

Sources: GitHub Copilot CLI documentation (June 2026), Microsoft internal memo coverage (The Verge / Forbes / Windows Central, May–June 2026), Anthropic Claude Code docs, Microsoft Foundry Claude Fable 5 docs (June 2026).