Memo

Memo MCP is an MCP server that saves and retrieves AI conversation context. You can use it to hand off conversations between different AI agents like Claude Code, Gemini CLI, and Codex.

You can start a session on a laptop and continue it on a desktop machine. The agent saves structured context like goals, completed tasks, pending tasks, decisions, and relevant files when you issue a specific command. The system returns a short ID. You use this ID anywhere to restore the exact context.

Features

• Saves AI conversation context into Upstash Redis.
• Retrieves previous conversation states using a unique short ID.
• Stores structured snapshots containing goals, completed tasks, pending tasks, key decisions, and relevant file paths.
• Encrypts stored data at rest and in transit.
• Expires data automatically after 24 hours by default.
• Accepts custom expiration times via a command-line flag.
• Supports self hosting on custom infrastructure using Vercel.

Use Cases

• Transfer an active debugging session from Claude to Cursor to utilize a different AI model.
• Pause a complex refactoring task at the end of the workday and resume it the next morning.
• Migrate an ongoing development conversation from a local laptop to a remote desktop workstation.
• Share a specific project state with another developer by passing the generated context ID.
• Host the API and storage on private infrastructure to maintain strict data ownership.

Installation

Claude Code

claude mcp add memo -- npx -y @upstash/memo

The -y flag tells npx to accept the installation prompt automatically.

OpenCode

{
  "mcp": {
    "memo": {
      "type": "local",
      "command": ["npx", "-y", "@upstash/memo"]
    }
  }
}

Claude Desktop and Cursor

{
  "mcpServers": {
    "memo": {
      "command": "npx",
      "args": ["-y", "@upstash/memo"]
    }
  }
}

Basic Usage

Type memo set in your AI chat interface to save the current context. The AI summarizes the conversation and returns an ID string like 4tJ630XqhCV5gQelx98pu. The agent stores a structured snapshot containing the goal, completed tasks, pending tasks, key decisions, and relevant file paths. This keeps restored conversations focused. It avoids reloading raw chat history.

Type memo get <id> to restore a previous context. Replace <id> with your specific string. The AI loads the previous context and continues the conversation.

Self-Hosting Instructions

You can self host the API and storage on your own infrastructure. The repository includes both the MCP server and the API.

Create a .env file in the project root to set up your environment. Add your Upstash Redis credentials to this file:

UPSTASH_REDIS_REST_URL=your-redis-url
UPSTASH_REDIS_REST_TOKEN=your-redis-token

Run this command to start the local development server:

npx vercel dev

Run this command to deploy the API to Vercel:

vercel

You must set the exact same environment variables in your Vercel project settings.

Modify the index.ts file to point the MCP server to your custom API. The default server uses https://memo-upstash.vercel.app. Update these variables in index.ts:

const GET_URL = "https://your-api.vercel.app/api/get";
const SET_URL = "https://your-api.vercel.app/api/set";

Configuration Options

The server accepts specific arguments to modify its behavior. The project operates under the MIT License.

• --ttl-mins: Sets the expiration time for stored context data in minutes. The default value is 1440. You can pass this argument in your configuration file. Example usage: "args": ["-y", "@upstash/memo", "--ttl-mins", "4320"].

FAQs

Q: How long are my saved contexts stored?
A: By default, contexts expire after 24 hours. You can change this with the --ttl-mins option when starting the server.

Q: Can I use my own Redis instance?
A: Yes, self‑hosting allows you to use your own Upstash Redis instance by setting the appropriate environment variables.

Q: What happens if I lose the ID?
A: The ID is the only way to retrieve a saved context. If you lose it, the context cannot be restored. Make sure to copy the ID after running memo set.

Q: Is my conversation data encrypted?
A: Yes, data stored in Upstash Redis is encrypted at rest and in transit. If you self‑host, encryption depends on your infrastructure.

Q: Can I save and restore context across different AI agents?
A: Yes, as long as both agents have the Memo MCP server installed. Save in one, restore in another.