# Layven

> A shared drive that AI agents mount over MCP. Claude, ChatGPT, Cursor, Claude Code and custom scripts read and write the same files. Every write is stored as its own version and attributed to the specific agent that made it, and any file or the whole tree can be restored to any earlier moment.

Layven exists because agents work at machine speed and make mistakes at machine speed. Storage that keeps every version and names the writer makes those mistakes recoverable instead of fatal. It is operated by Trueguard OÜ, an Estonian company, with all file content stored in France.

## What it is

- Shared file storage that any MCP client mounts with one config block. No SDK, no special agent code.
- Every write creates a new version, attributed to the agent token that made it. Nothing is overwritten in place.
- Any file restores to any earlier version; any folder restores to any earlier point in time. The restore is itself a new version, so undo is undoable.
- An activity feed records every read, write, move, delete, restore and lock, with the agent responsible.
- Advisory locks and optimistic concurrency for agents that would otherwise clobber each other.
- Per-token permissions: each agent token is scoped to specific workspaces, at read-only, read-write or admin level, with optional path prefix allow and deny rules.

## Connecting an agent

- MCP endpoint: https://api.layven.io/mcp (Streamable HTTP)
- Authentication: an `Authorization: Bearer` header with a Layven agent token, or OAuth for clients such as the Claude.ai connector that offer no custom header field.
- Claude Code: `claude mcp add --transport http layven https://api.layven.io/mcp --header "Authorization: Bearer <token>"`
- Cursor and other config-file clients: add an entry pointing at the same URL with the same header.
- Tools exposed: list, read, write, edit, request_upload, finalize_upload, move, delete, history, grep, restore, lock, unlock, activity.

## Pricing

- Flat monthly pricing with no metering and no credits. Agent request volume does not change the bill.
- Free: 5 GB, 1 workspace, 2 agent tokens, 1 member, 30 days of version history.
- Pro, $19/month: 100 GB, 5 workspaces, 25 agent tokens, 3 members, 1 year of version history.
- Scale, $49/month: 500 GB, 25 workspaces, 100 agent tokens, 10 members, 1 year of version history.
- Enterprise: custom, by agreement.

## Data handling

- All file content is stored in Gravelines, France, encrypted at rest.
- Product analytics runs on PostHog EU Cloud (Frankfurt) via a first-party domain, and never receives file content, file names or the activity log. No advertising, ad networks or data brokers.
- Customer content is never sent to any third-party AI or embedding provider, and is never used to train models.
- Files export as plain files.

## Links

- [Layven](https://layven.io): what it is, how it works, pricing and FAQ.
- [Knowledge base](https://layven.io/knowledge-base): step-by-step guides for connecting Claude, ChatGPT, Gemini, Claude Code, Cursor and any other MCP client.
- [Google Drive MCP](https://layven.io/knowledge-base/google-drive-mcp): the three servers called "Google Drive MCP", what each can do to a file, and where Drive's revision retention, actor attribution and per-file restore fall short for agent working storage.
- [Dropbox MCP](https://layven.io/knowledge-base/dropbox-mcp): Dropbox's own remote MCP server, its twenty-five tools including a real folder restore, and the limits that decide whether it can be an agent's working directory: a 5 MB ceiling on reads and writes, whole-file rewrites with no patch tool, plan-tiered version history, full-text search gated to Business plans, and attribution that names the app rather than the agent.
- [Privacy Policy](https://layven.io/privacy): what data is held, where it lives, and how deletion works.
- [Terms of Service](https://layven.io/terms): the agreement, acceptable use and liability.
- [Sub-processors](https://layven.io/subprocessors): every third party that touches Layven data.