CLAUDE.md vs .cursorrules: What's the Difference?

If you've used Cursor, you already understand the core idea behind CLAUDE.md. Both are plain-text files that tell an AI how to work with you — your stack, your conventions, your preferences — so you stop re-explaining yourself every conversation. The concepts are close cousins. The details differ in ways that matter when you move between the two. This is a straight comparison: what each file is, where it lives, how they differ, and whether the one you already wrote can be reused.

The short version

.cursorrules (and the newer Cursor Rules in .cursor/rules/) is Cursor's mechanism for giving its AI persistent project instructions. CLAUDE.md is the equivalent for Claude — read by Claude Code, the Claude apps, and Cowork mode as standing context. Same job, different tools. The biggest practical difference: Claude reads your CLAUDE.md before every conversation, so length is a real cost — the working ceiling is under ~2,000 tokens. Cursor's newer rules system is more granular (multiple scoped rule files, auto-attached by file globs).

Side by side

What they have in common

Everything that makes a .cursorrules good makes a CLAUDE.md good. Name your stack explicitly. State preferences as specific executable rules, not adjectives. Include an explicit "never do this" list. Keep it scannable. The principles in /guide/how-to-write-claude-md map one-to-one onto Cursor Rules and vice versa.

Where they differ — and what to watch for

1. Length matters more for CLAUDE.md. Claude loads the whole file every conversation, so bloat is a tax paid every turn. If porting a long .cursorrules, /guide/optimize-claude-md to trim to essentials.
2. CLAUDE.md is one file; Cursor Rules can be many. Cursor splits rules into .mdc files auto-attached by glob; CLAUDE.md is single global. When migrating, consolidate the durable always-true rules and drop hyper-specific per-directory rules.
3. CLAUDE.md isn't just for code. Cowork mode reads it to run a job search, manage consulting clients, or handle email. See /guide/what-is-a-claude-md-file and /guide/claude-cowork-guide.
4. Boundaries translate directly. The "never do this" list is the most valuable section and ports verbatim.

Can I reuse my .cursorrules as a CLAUDE.md?

Mostly yes, with two edits: consolidate scoped rule files into one, and trim for tokens (under ~2,000). Rename to CLAUDE.md, drop in project root. Fastest path: paste your role + rules into the free Builder — it generates a clean CLAUDE.md with a Claude Fluency Score (clarity, specificity, format, voice, boundaries, token count). Already running it? Paste a real chat into /analyzer.

FAQ

Does Claude read .cursorrules?

No. Claude reads CLAUDE.md (and ~/.claude/), not Cursor's files. Similar ideas, but you must provide instructions in a CLAUDE.md.

What is the CLAUDE.md equivalent of Cursor Rules?

CLAUDE.md is the equivalent — Claude's persistent project-context file. It's a single file loaded every conversation (keep under ~2,000 tokens) and works for non-coding tasks too.

How do I convert my .cursorrules to a CLAUDE.md?

Merge scoped files into one, cut anything that won't change behavior, keep under ~2,000 tokens, save as CLAUDE.md in project root. Or use the /generator Builder.

Which is better, CLAUDE.md or .cursorrules?

Neither — different tools. Cursor Rules give finer per-file targeting in the Cursor IDE; CLAUDE.md gives one portable context file across Claude Code, the apps, and Cowork, including non-coding work.

Do I have to keep my CLAUDE.md short?

Effectively yes — read before every conversation, ceiling ~2,000 tokens. See /guide/how-long-should-claude-md-be and /guide/claude-md-checklist.