# MemScope vs CLAUDE.md and AGENTS.md files in git

Instruction files in your repository and a shared memory with per-person access solve different problems. Here is when each fits.

## What CLAUDE.md and AGENTS.md files are

CLAUDE.md files are markdown files you write to give Claude Code persistent instructions for a project, your personal workflow or your organization; Claude Code reads them at the start of every session, and can read a repository’s AGENTS.md in place of CLAUDE.md. A project file is shared with the team through git, versioned with the code, and needs nothing to install. Anthropic’s docs describe them as the place for build commands, conventions, project layout and “always do X” rules.

## Side by side

| | CLAUDE.md and AGENTS.md files | MemScope |
| --- | --- | --- |
| Per-person access | A project CLAUDE.md is team-shared (everyone with the repository). CLAUDE.local.md is only you. No per-person access inside a shared file is documented. | Yes. Each person’s agent recalls only memories that person is permitted to see; a deterministic policy decides. |
| Works across different AI tools | Claude Code reads CLAUDE.md, and can read a repository’s AGENTS.md in place of it. Whether other tools read CLAUDE.md: not documented there. | Yes. Claude Code, Cursor, ChatGPT and other MCP clients share one memory. |
| Sensitivity levels | Not documented. | Four levels: Public, Internal, Restricted, Highly restricted. Higher levels need more access. |
| Audit trail | Git history shows who changed a file and when (a git feature). No separate audit trail is documented. | Yes. Every change is audited. |
| Who writes the content | You write CLAUDE.md. Auto memory is separate: notes Claude writes itself, kept on your machine and not shared across machines. | People and their agents save explicit memories; AI only suggests labels. |
| Cost | No separate charge documented; the files live in your repository. | $15 per connected user per month after a 60-day trial with 3 connected users. |

## When CLAUDE.md and AGENTS.md files are enough

- Everyone who can read the repository may read the instruction, which is true of build commands, code style, project layout and “always do X” rules.
- You want the instruction reviewed in a pull request and versioned with the code it describes.
- Your team uses one AI tool, or tools that read the same file.

In these cases a file in the repository is the right answer. Use it, keep it short, and skip MemScope.

## When you need MemScope

- Some knowledge is for one team or project, and should not reach everyone who can read the repository.
- Developers use different AI tools and you want one memory they all read and write.
- You need each change to who can see a memory recorded in an audit log.

- [Start free](https://memscope.ai/signup)

## Source and date

Statements about CLAUDE.md and AGENTS.md files come from Claude Code docs: How Claude remembers your project, checked November 2026. “Not documented” means those docs do not say. Products change; check the source. https://code.claude.com/docs/en/memory

---

Web page: https://memscope.ai/compare/claude-md · All facts for AI agents: https://memscope.ai/llms.txt
