# fimemory [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/FutureAiIndustries/FI-Memory  
**GitHub Stars:** 0  
**npm Downloads (last month):** 216  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/fimemory

## Description
Local-first shared memory for AI tools. Your files, encrypted at rest by default; budgeted reads.

## Claude Desktop Quick Installation
Heuristic fallback — verify the package name and runner against the repository README before running it. Uses `npx` (confidence: low):

```json
"mcpServers": {
  "fimemory": {
    "command": "npx",
    "args": ["-y","fimemory"],
    "env": {
      "GESTALT_PASSPHRASE": ""
    }
  }
}
```

**Requires environment variables:** `GESTALT_PASSPHRASE` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What fimemory MCP server does

The fimemory MCP server gives AI tools access to a shared memory store kept on your own computer. The store is an ordinary folder of files rather than a proprietary database, and nothing is uploaded by default. New stores use encryption at rest unless setup is run with `--plaintext`; an encrypted store can later be exported to readable Markdown after unlocking.

The store is intended for notes, project context, and other information that agents may need across sessions. Agents only use the memory when they are connected and instructed to consult it. The package cannot force a host application to retrieve memory on every turn.

## How it works

Running `fimemory setup` creates a store when one does not already exist, installs the MCP entry for detected hosts, adds memory rules, installs the Claude Code retrieval hook when applicable, and runs a diagnostic check. Each step reports its own result, so a failure in one configuration does not prevent later checks. Repeating setup is safe; completed work is reported as unchanged.

Encrypted setup asks for a passphrase and prints a 24-word recovery phrase once. The store remains unlocked locally for roughly eight hours. Scripted use can provide `--passphrase`, set `GESTALT_PASSPHRASE`, or explicitly select plaintext mode. Losing both the passphrase and recovery phrase makes the data unrecoverable.

The fimemory MCP server can be used by Claude Code, Codex CLI, Gemini CLI, Grok CLI, Windsurf, Cursor, and Claude Desktop configurations described by the project. Claude Code receives a command to paste because its CLI owns the relevant configuration schema. Retrieval behavior differs by host: Claude Code can use a hook, while other hosts rely on the installed rules and MCP access.

## Setup and configuration

Install the package globally and run the guided setup:

```bash
npm i -g fimemory
fimemory setup
```

Use `fimemory onboard` after setup to approve an initial suggested edit, answer questions about working habits, and verify a search against the resulting store. Use `fimemory doctor` to inspect the installation and identify missing configuration. Individual commands are also available for initialization, MCP installation, hooks, and rules.

For an unencrypted store, run `fimemory setup --plaintext`. A plaintext store can later be converted with `fimemory encrypt`. If npm cannot write to its global directory, configure a user-owned prefix instead of using elevated permissions.

## Tools and capabilities

- Connects multiple AI hosts to one local memory folder.
- Stores information as user-owned files.
- Encrypts new stores by default and supports passphrase-based recovery.
- Installs MCP configuration, host rules, and a Claude Code retrieval hook where supported.
- Provides `setup`, `onboard`, `doctor`, `encrypt`, `join`, and `pull` commands.
- Supports movement between machines through a Git remote that the user controls.

## Limitations and notes

Automated synchronization is not included. Multi-machine use requires configuring a private Git remote, running `fimemory join <url>` on another machine, and using `fimemory pull` before and after sessions. Conflicting edits do not silently merge; the losing version becomes a pending proposal for review.

The project reports hands-on use on Windows, macOS, and Linux, but host configuration paths for Windsurf, Cursor, and Claude Desktop are vendor-documented and not verified on disk by the project. Linux clipboard operations require `xclip` or `wl-clipboard` when applicable. Team features beyond a user's own remote are outside the shipped scope.

_Full upstream README: https://allmcps.com/mcp/fimemory/readme_