Memorize

Memory & Workflow Management for AI

A complete MCP toolkit with 3 powerful tools: save memories as JSON, sync bidirectionally with Supabase Cloud, and pull workflow instructions—empowering AI agents to remember and collaborate.

How to get started

Set up Memorize MCP in minutes and give your AI assistant a persistent memory across all your devices.

1Step 1

1. Install Dependencies

Clone the repository and install dependencies using Bun runtime (v1.2.x or higher).

git clone https://github.com/congthien2003/sabo-mcp.git
cd sabo-mcp
bun install
2Step 2

2. Setup Supabase (Optional)

Create a Supabase project, run the SQL migration from docs/version1.1/migrations/001_initial_schema.sql, and get your credentials.

# Required environment variables:
MEMORIZE_MCP_SUPABASE_URL
MEMORIZE_MCP_SUPABASE_SERVICE_ROLE_KEY
MEMORIZE_MCP_PROJECT_SLUG

# Optional:
MEMORIZE_MCP_PROJECT_ROOT
MEMORIZE_MCP_TARGET_PROJECT_DIR
3Step 3

3. Configure MCP Client

Add memorize-mcp to your MCP client config (Claude Desktop, VS Code, etc.) using stdio transport.

{
  "memorize-mcp": {
    "command": "bun",
    "args": ["run", "path/to/sabo-mcp/index.ts"],
    "env": {
      "MEMORIZE_MCP_PROJECT_ROOT": "C:/memories",
      "MEMORIZE_MCP_SUPABASE_URL": "https://xxx.supabase.co",
      "MEMORIZE_MCP_SUPABASE_SERVICE_ROLE_KEY": "your-key",
      "MEMORIZE_MCP_PROJECT_SLUG": "my-project",
      "MEMORIZE_MCP_TARGET_PROJECT_DIR": "C:/your-project"
    }
  }
}
4Step 4

4. Use the 3 Tools

Start with pull_workflows to get instructions, sync_memorize to pull existing memories, then save_memorize after completing tasks.

# pull_workflows: Download workflow instructions
# sync_memorize: Sync memories from cloud
# save_memorize: Save new work summaries

# All tools available via MCP client interface

Frequently Asked Questions

Everything you need to know about memorize-mcp v1.2.1 and its three powerful MCP tools

What is memorize-mcp?
Memorize MCP is a lightweight Model Context Protocol (MCP) server that provides 3 powerful tools: save_memorize (v1.0) to save memories to JSON + cloud, sync_memorize (v1.2) to pull memories from cloud, and pull_workflows (v1.2.1) to download workflow instructions for AI agents.
What's the difference between save_memorize and sync_memorize?
save_memorize uploads: it saves work summaries from local to JSON file and syncs UP to cloud. sync_memorize downloads: it pulls memories FROM cloud down to local storage, with smart timestamp comparison to avoid overwriting newer local files. Together they provide bidirectional sync.
What is the pull_workflows tool (v1.2.1)?
Introduced in v1.2.1, pull_workflows downloads a .workflows folder from a source (local/Supabase/GitHub) to your project directory. These workflow files contain instructions that guide AI agents on how to perform standardized tasks, ensuring consistency across projects and team members.
Can I use this across multiple devices?
Yes! Configure the same Supabase credentials and project slug on all devices. Use save_memorize on device A to upload memories, then sync_memorize on device B to download them. pull_workflows ensures all devices have the same workflow instructions for consistent AI behavior.
How does the offline-first approach work?
Every save_memorize call always writes to a local JSON file first, ensuring your memory is preserved even without internet. If Supabase is configured, the tool attempts to sync to cloud afterward. If cloud sync fails, your local file remains intact—no data is lost.
Do I need Supabase to use memorize-mcp?
No. Supabase is optional. Without it, memorize-mcp works as a purely local memory storage solution, saving files to the directory specified by MEMORIZE_MCP_PROJECT_ROOT (default: ./.memories/data). pull_workflows can also work with local source files.
How are memories organized?
Memories are saved as JSON files with a filename you specify. Each file includes topic, timestamp, content, and createdAt fields. You can organize them by project using the projectSlug parameter to isolate memories for different codebases or workspaces.
Which MCP clients are supported?
Memorize MCP works with any MCP client that supports stdio transport, including Claude Desktop, VS Code extensions, or custom integrations. You just need to configure the client to run the server with Bun runtime (v1.2.x or higher).

Ready to save your AI's memory?

Start using memorize-mcp today to give your LLM persistent context across sessions and devices with offline-first JSON storage and optional cloud sync.