remote-memory-mcp-server

Remote Memory MCP Server

A GitHub-integrated remote memory management MCP server that syncs knowledge graph data with GitHub repositories for remote storage and collaboration.

Features

• CRUD operations for entities, relations, and observations

• Real-time synchronization with GitHub repositories

• Conflict detection and resolution

• Automatic/manual synchronization options

• Search and filtering capabilities

• Project-level memory isolation (v1.4.0)

  • Multiple projects, each with independent memory

  • Persistent active project (stored in GitHub memory/index.json)

  • Per-call project override without switching active project

• Enhanced entity query features (v1.3.0)

  • Entity list retrieval (with filtering, sorting, pagination)

  • Quick entity name lookup

  • Entity type statistics

  • Date range filtering

• Enhanced commit messages (customizable)

• Backup functionality (per-project)

• Commit history tracking

• Optional auto-push (AUTO_PUSH environment variable)

• Local mirror mode (v2): persist the active project's graph to a local JSONL file in anthropic-memory canonical format, so external tools (e.g. graph-view) can read and write it directly. Includes divergence guard for multi-PC safety.

Related MCP server: Project MCP Server

Installation

cd C:\YOUR_PATH\remote-memory-mcp
npm install
npm run build

Configuration

Required Parameters

• GITHUB_TOKEN: GitHub Personal Access Token (requires repo permissions)

• GITHUB_OWNER: GitHub repository owner

• GITHUB_REPO: GitHub repository name

Optional Parameters

• GITHUB_BRANCH: Branch name to use (default: main)

• SYNC_INTERVAL: Auto-sync interval in seconds (0 for manual)

• AUTO_PUSH: Auto-push after CRUD operations (true/false, default: false)

• PROJECT_NAME: Active project on startup (default: from memory/index.json, fallback: "default")

• LOCAL_MIRROR_PATH: Absolute path to a local JSONL mirror file (v2). When set, every mutation of the active project is mirrored to this file in anthropic-memory canonical format. External tools (e.g. graph-view) can read and write the same file. Unset = legacy behavior (in-memory + GitHub only).

Claude Desktop Setup

Add to your claude_desktop_config.json file:

{
  "mcpServers": {
    "remote-memory": {
      "command": "node",
      "args": ["C://YOUR_PATH//remote-memory-mcp//dist//index.js"],
      "env": {
        "GITHUB_TOKEN": "YOUR_GITHUB_TOKEN_HERE",
        "GITHUB_OWNER": "YOUR_GITHUB_USERNAME",
        "GITHUB_REPO": "YOUR_GITHUB_REPO",
        "GITHUB_BRANCH": "main",
        "SYNC_INTERVAL": "0",
        "AUTO_PUSH": "false",
        "PROJECT_NAME": "my-project"
      }
    }
  }
}

Usage

For detailed API usage and examples, see SPEC.md.

Project-level Memory (v1.4.0)

Each project stores memory independently in the GitHub repository:

memory/
├── index.json           ← project index + active project pointer
├── graph.json           ← "default" project (backward compatible)
├── blog/
│   └── graph.json       ← "blog" project
└── my-app/
    └── graph.json       ← "my-app" project

Quick Start

// 1. Create a project
create_project({ name: "blog", description: "Blog memory" })

// 2. Switch to it
switch_project({ project: "blog" })

// 3. Work normally — all tools now target "blog"
create_entities({ entities: [...] })

// 4. Access another project without switching
read_graph({ project: "my-app" })

Active Project Priority

PROJECT_NAME env var → memory/index.json → "default"

Local Mirror Mode (v2)

Set LOCAL_MIRROR_PATH to enable mirroring the active project's graph to a local JSONL file in anthropic-memory canonical format. External tools (e.g. graph-view) can read and write the same file, giving the user a UI on top of a remote-memory backed graph without graph-view needing to know about GitHub.

Behavior

• Bootstrap: if the mirror file exists and its sidecar (<mirror>.sync-state.json) names the current active project, remote-memory loads from the mirror (preferring the user's local edits over GitHub). Otherwise, GitHub is pulled and the mirror is seeded.

• Before every tool call: mirror mtime is checked; if it changed (external writer), the in-memory graph is reloaded from the mirror.

• After every mutation: the in-memory graph is atomically written back to the mirror (.tmp + rename). If the file was modified externally during the operation (race), the in-memory mutation is rolled back and an error is surfaced.

• JSONL line format: anthropic memory MCP compatible — {"type":"entity"|"relation", ...} with optional createdAt/updatedAt extension fields. Unknown fields are silently dropped on read.

• Per-call project override (e.g. read_graph({ project: "blog" })) does not touch the mirror — the mirror always represents the active project.

• switch_project rewrites the mirror with the new active project's graph and updates the sidecar.

Divergence Guard (multi-PC safety)

When the same GitHub repo is shared across multiple machines, sync_pull / sync_push follow this policy (active project only):

sync_push applies the same baseline check — if GitHub advanced since the last sync, the push is refused (remote-ahead). force_sync is the escape hatch and bypasses both guards.

The baseline (last-pull SHA + graph digest + project name) is persisted to the sidecar file (<LOCAL_MIRROR_PATH>.sync-state.json) so it survives process restarts. The sidecar is owned by remote-memory; external tools must not modify it.

graph-view integration example

{
  "mcpServers": {
    "remote-memory": {
      "command": "node",
      "args": ["C:/YOUR_PATH/remote-memory-mcp/dist/index.js"],
      "env": {
        "GITHUB_TOKEN": "ghp_...",
        "GITHUB_OWNER": "...",
        "GITHUB_REPO": "...",
        "LOCAL_MIRROR_PATH": "D:/memory/memory.jsonl"
      }
    },
    "graph-view": {
      "command": "node",
      "args": ["D:/mcpapps/graph-view/dist/server.js"],
      "env": {
        "MEMORY_FILE_PATH": "D:/memory/memory.jsonl"
      }
    }
  }
}

graph-view auto-detects LOCAL_MIRROR_PATH (via env or mcpServers.remote-memory.env) and switches to mirror backend automatically.

Data Structure

Memory data is stored per project in the GitHub repository:

{
  "entities": {
    "Kim Kim": {
      "name": "Kim Kim",
      "entityType": "Person",
      "observations": ["Software developer", "Lives in Seoul"],
      "createdAt": "2025-01-01T00:00:00.000Z",
      "updatedAt": "2025-01-01T00:00:00.000Z"
    }
  },
  "relations": [
    {
      "from": "Kim Kim",
      "to": "Company ABC",
      "relationType": "works_at",
      "createdAt": "2025-01-01T00:00:00.000Z"
    }
  ],
  "metadata": {
    "version": "1.0.0",
    "lastModified": "2025-01-01T00:00:00.000Z",
    "lastSync": "2025-01-01T00:00:00.000Z"
  }
}

Architecture

Core Components

1. GitHubClient: Handles GitHub API interactions

2. MemoryGraphManager: Manages the in-memory knowledge graph

3. SyncManager: Handles synchronization and project management

4. RemoteMemoryMCPServer: Main MCP server class

Synchronization Strategy

1. Conflict Resolution: Prioritizes based on latest modification timestamp

2. Auto-push: Immediately pushes local changes to remote

3. Auto-pull: Checks for remote changes at configured intervals

4. Force Sync: Performs bidirectional sync ignoring conflicts

Important Notes

• Requires GitHub Personal Access Token (with repo permissions)

• GitHub API limits: 5,000 requests per hour for authenticated users

• Network connection required

• Project names: alphanumeric, hyphens, underscores only; "default" is reserved

License

MIT License - Free to use, modify, and distribute

Changelog

v1.4.0

• Project-level memory isolation

  • list_projects: List all projects and active project

  • create_project: Create a new isolated project

  • switch_project: Switch active project (persisted to GitHub)

• project parameter on all tools: Target any project per-call without switching

• Per-project backup paths: backups/{project}/backup-*.json

• Added PROJECT_NAME environment variable

• Server version bumped to 1.4.0

v1.3.0

• New query tools

  • list_entities: Retrieve entity list (with filtering, sorting, pagination)

  • get_entity_names: Quick entity name lookup

  • get_entity_types: Entity type statistics

• Enhanced query capabilities

  • EntityType filtering

  • Date range filtering (based on createdAt)

  • Sort options (createdAt, updatedAt, name)

  • Pagination (limit, offset)

• Improved handling of large datasets

v1.2.0

• Prevented unnecessary auto-commits on initialization

• Added AUTO_PUSH environment variable for optional auto-push

• Added logic to prevent pushing empty graphs

• Improved initial load state tracking

v1.1.0

• Custom commit message support

• Added backup system (create_backup)

• Commit history tracking (get_commit_history)

• Automatic commit message generation

v1.0.0

• Initial release