# agentleFS

> agentleFS is a permissioned, versioned store of files that people and their AI agents use together over MCP. Each agent reads only what the person behind it can read.

Agent permissions for the files your team shares. One folder Claude Code, Cursor and Codex all read, where an agent reads only what its person can, every write is a version, and a folder can be shared outside your company.

agentleFS is a permissioned file store that AI agents read over MCP (Model
Context Protocol). Access is decided per PERSON rather than per API key: an
agent connected on someone's behalf reads exactly what that person is allowed
to read, and the same authorization filter runs in the SQL WHERE clause before
ranking, so content the caller cannot reach is never returned, listed or matched.

## Pages

- [Agent permissions for files your team shares — agentleFS](https://agentlefs.com/): One folder Claude Code, Cursor and Codex all read over MCP. An agent reads only what its person can, every write is a version, and folders share outward.
- [CLAUDE.md vs AGENTS.md — agentleFS](https://agentlefs.com/context/agents-md-vs-claude-md): Compare CLAUDE.md and AGENTS.md with counts collected from public GitHub projects. Browse examples of both in the agentleFS instruction corpus.
- [Blog: CLAUDE.md, AGENTS.md and agent context — agentleFS](https://agentlefs.com/blog): What thousands of CLAUDE.md and AGENTS.md files on public GitHub show, measured from their git history, and how to write the files your agents read.
- [CLAUDE.md best practices from 5,786 real files — agentleFS](https://agentlefs.com/blog/claude-md-best-practices): CLAUDE.md and AGENTS.md best practices from 5,786 real files: what people cut as models improved, what they kept, and what to enforce with hooks.
- [Documentation — agentleFS](https://agentlefs.com/docs): Connect an MCP client to agentleFS, then read how permissions, versions and organizations decide what an agent can reach.
- [Questions about agent access and sharing — agentleFS](https://agentlefs.com/docs/faq): Answers people look for before and after connecting an agent: how search works, who can read a folder, and how sharing works inside your organization.
- [Folders and version history — agentleFS](https://agentlefs.com/docs/folders): How agentleFS stores content in folders and files, keeps every document's own version history, and lets permissions inherit down the folder tree.
- [People, agents, tokens and grants — agentleFS](https://agentlefs.com/docs/identity): A token authenticates a person or an agent and carries no access of its own. What it can read comes from grants that are checked on every request.
- [Connect Claude or Cursor over MCP — agentleFS](https://agentlefs.com/docs/connect): Add agentleFS to Claude Code, Claude Desktop, Cursor or any MCP client. You sign in through the browser, and the agent reads what you are allowed to read.
- [REST API for documents and search — agentleFS](https://agentlefs.com/docs/api): Call agentleFS from your own code over HTTP. Read, write, move and search documents with an API key, under the same permissions an agent receives.
- [Console reference: files, access, audit — agentleFS](https://agentlefs.com/docs/reference): What each part of the agentleFS console does: browsing and editing files, granting access to people and agents, mirroring sources, and the audit log.
- [Privacy policy — agentleFS](https://agentlefs.com/privacy): What agentleFS stores, how long it is kept, and which third parties process it. Includes how Google user data from connected accounts is handled.
- [Support — agentleFS](https://agentlefs.com/support): How to get help with agentleFS: the address that reaches us, the docs that answer the most common questions, and where to file a plugin issue.
- [Terms of service — agentleFS](https://agentlefs.com/terms): The terms covering an agentleFS account, the content you upload to it, and the agents you connect to it on your organization's behalf.

## For agents

- [Connect an agent to agentleFS](https://agentlefs.com/connect.md): Install instructions written for the agent itself. It runs the install for Claude Code, Codex or Cursor, then hands the browser sign-in to the person.

## Contact

Write to contact@agentlefs.com. A person reads it.

## Questions and answers

The full FAQ is inline at [llms-full.txt](https://agentlefs.com/llms-full.txt), 51 entries
covering connecting a client, the permission model, sharing inside your
organization, and what happens to your content.