# MemoFS MCP Server [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/memo-fs/memofs  
**GitHub Stars:** 6  
**npm Downloads (last month):** 53  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/memofs-mcp-server

## Description
Persistent memory and virtual filesystem MCP server for AI agents.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "memofs-mcp-server": {
    "command": "npx",
    "args": ["-y","@memofs/cli"]
  }
}
```

## Documentation

## What MemoFS MCP Server MCP server does

MemoFS MCP Server MCP server exposes the MemoFS memory runtime to AI agents through the Model Context Protocol. Its storage model is file-first: project memory is kept beneath `.memofs/` rather than hidden inside a database. Markdown files hold durable facts and notes, while JSONL and JSON files record events, graph data, manifests, archives, and restore snapshots.

The runtime supports project-wide core memory, timestamped notes, recall, virtual paths, snapshots, and restore operations. It can also maintain relationships between extracted entities and preserve action or session information described by the broader MemoFS runtime. The repository labels the project beta.

## How it works

A local filesystem store is the default operating mode. MemoFS resolves settings from constructor options, environment variables, and `.memofs/config.json`, in that order. The documented modes are `local`, which uses only the filesystem; `hybrid`, which combines local storage with cloud synchronization policies; and `memory`, which keeps data volatile for tests.

Recall works offline with lexical BM25 and fuzzy matching. Applications can add semantic retrieval through OpenAI or Voyage AI adapters, or use the Transformers.js ONNX adapter for in-process vector search without an API key. Memory can be anchored to project files, schemas, byte hashes, and symbol paths, with drift detection affecting stale knowledge during queries.

## Setup and configuration

For an existing AI-agent project, run `npx @memofs/cli init`. The CLI creates the `.memofs/` directory, adds project rules, prepares platform lifecycle hooks, and configures the local MCP server. The README lists Cursor, Claude Code, Codex, Copilot, and Cline as supported agent environments for this workflow.

Developers embedding the runtime in a Node.js or TypeScript application can install `@memofs/core` and create a filesystem-backed store with `createNodeFsMemoryStore`. The runtime is then initialized with a project identifier and a selected mode. The repository also publishes `@memofs/mcp-server` as the MCP server package.

## Tools and capabilities

The documented runtime operations include:

- Reading and writing core memory and notes.
- Recording durable notes with content and a memory kind.
- Recalling relevant stored information.
- Creating and restoring versioned snapshots.
- Using virtual AgentFS paths with lease-locking.
- Synchronizing local data with MemoFS Cloud in hybrid deployments.
- Running a memory linter for broken references, contradictions, and provenance links.
- Tracking causal lineage through action receipts and task references.

## Limitations and notes

MemoFS MCP Server MCP server is local-first and does not require a cloud account for local operation. Team workspaces, access control, a memory dashboard, and a hosted managed MCP endpoint belong to MemoFS Cloud rather than the open-source local setup. Cloud synchronization is available as a client capability, while hosted services are separate offerings.

Semantic search is not the default described path: offline lexical and fuzzy recall works without configuration, while vector search requires a selected adapter. The source identifies the project as beta, so developers should evaluate its behavior against their agent workflow before adopting it for production-critical memory.

_Full upstream README: https://allmcps.com/mcp/memofs-mcp-server/readme_