Cursor Checkpoints vs Git vs mrq: Which Should You Use?
Comparing version control options for AI coding. When to use Cursor's built-in checkpoints, Git, or automatic snapshot tools.
When you’re coding with Cursor, you have three main options for protecting your work: Cursor’s built-in checkpoints, traditional Git, or automatic snapshot tools like mrq. Each serves a different purpose, and understanding when to use each will save you from losing work.
Cursor Checkpoints
Cursor creates a checkpoint automatically every time you send a prompt in Composer. You can see these as “Checkpoint created” messages, and clicking “Restore” reverts all changes from that prompt.
This is convenient for quick undo. You try something, it doesn’t work, you click Restore and you’re back. No commands to remember, no decisions to make.
But checkpoints have significant limitations. They only cover Composer interactions, so Tab completions aren’t protected. They’re session-based, so they don’t persist after you restart Cursor. They revert everything from a prompt, so if you made manual edits after accepting AI changes, those edits are lost too.
Checkpoints are good for “let me try this and maybe undo immediately.” They’re not a version control strategy.
Git
Git is the industry standard for a reason. It provides permanent history, works with any editor or AI tool, and enables collaboration. Every professional development project uses Git.
For AI coding, Git works well if you’re disciplined about committing before AI interactions. The workflow is straightforward: commit, prompt the AI, review the diff, commit again if it’s good, revert if it’s not.
The problem is that discipline breaks down when you’re iterating quickly. You’re in flow, things are working, you want to try “just one more thing” before committing. Then the AI goes haywire, and you realize you haven’t committed in twenty minutes.
Git protects what you committed. It can’t protect what you didn’t.
mrq (Automatic Snapshots)
mrq takes a different approach. Instead of requiring you to decide when to save, it captures every file change automatically. You run mrq watch once when you start working, and everything from that point forward is recorded.
mrq watch # Start protection
# ... code with Cursor, Claude Code, any AI tool ...
mrq history # See all snapshots
mrq restore abc123 # Go back to any point
Unlike Cursor checkpoints, mrq works across sessions and with any AI tool. Unlike Git, it doesn’t require you to remember to commit. It’s a safety net that’s always there.
The tradeoff is that automatic snapshots don’t create meaningful history the way Git commits do. You won’t have a log that says “added user authentication” and “fixed checkout bug.” You’ll have a stream of every change, which is great for recovery but not for understanding project evolution.
When to Use Each
Cursor Checkpoints are for immediate undo during a Composer session. You tried something, it didn’t work, you want to undo and try a different prompt. Click Restore and move on.
Git is for intentional milestones. You finished a feature, you’re about to deploy, you want to mark a meaningful point in your project’s history. git commit -m "User registration complete" captures that intent.
mrq is for continuous protection while you work. It runs silently, captures everything, and ensures you can always recover. It’s especially valuable during rapid iteration when you’re experimenting and not thinking about version control.
The Practical Answer: Use All Three
These tools aren’t mutually exclusive. They layer together well.
Run mrq in the background for automatic protection. Use Cursor checkpoints for quick undo during Composer sessions. Commit to Git when you hit milestones worth documenting.
# When starting work
mrq watch
# While working
# ... Cursor checkpoints for quick undo ...
# ... mrq capturing everything in the background ...
# When you hit a milestone
git add .
git commit -m "Feature complete"
This gives you the best of each: meaningful Git history for collaboration and deployment, Cursor’s convenient one-click undo for experimentation, and mrq’s continuous protection for everything in between.
You get to iterate as fast as AI allows, knowing you can always recover, while still building the kind of Git history you’d want to show a collaborator.
Related Reading
- How to Undo Changes in Cursor AI - Complete Cursor guide
- Best Practices for AI Pair Programming - Work effectively with AI
- Do I Need to Learn Git? - Git advice for AI coders
mrq is automatic versioning for AI coding. Works alongside Git and your editor’s built-in features.
Written by mrq team