内容
# Memo MCP
An MCP server that saves and retrieves AI conversation context. Hand off conversations between AI agents seamlessly. Free and no account required.
## What it does
When you're working with an AI agent and need to:
- **Switch agents** - Claude can't fix your bug? Try Cursor or Copilot with dense context
- **Continue later** - Save progress and pick up where you left off
- **Move machines** - Start on laptop, continue on desktop
Just say `memo set` and the agent will save structured context (goal, completed tasks, pending tasks, decisions, relevant files). Get a short ID back, use `memo get <id>` anywhere to restore context.
https://github.com/user-attachments/assets/f96798e1-2f8b-4de3-9431-b7bc52c58dfa
## Installation
### Claude Code
```bash
claude mcp add memo -- npx -y @upstash/memo
```
### OpenCode
Add to your `opencode.json`:
```json
{
"mcp": {
"memo": {
"type": "local",
"command": ["npx", "-y", "@upstash/memo"]
}
}
}
```
### Claude Desktop / Cursor
Add to your MCP config:
```json
{
"mcpServers": {
"memo": {
"command": "npx",
"args": ["-y", "@upstash/memo"]
}
}
}
```
## Usage
### Save context
```
memo set
```
The AI will summarize the conversation and return an ID like `4tJ630XqhCV5gQelx98pu`.
### Restore context
```
memo get 4tJ630XqhCV5gQelx98pu
```
The AI will load the previous context and continue where you left off.
## What gets saved
When you run `memo set`, the agent stores a structured snapshot, for example:
- Goal
- Completed tasks
- Pending tasks
- Key decisions
- Relevant files (paths only) or references
This keeps restored conversations focused and avoids reloading raw chat history.
## Security
Your conversation context is stored in Upstash Redis with encryption enabled at rest and in transit. Data expires automatically after 24 hours (configurable via `--ttl-mins`).
If you prefer full control over your data, you can self-host both the API and storage using your own infrastructure. See the Self-hosting section below.
## Self-hosting
The repo includes both the MCP server and the API. To self-host:
### 1. Set up environment
Create a `.env` file with your Upstash Redis credentials:
```
UPSTASH_REDIS_REST_URL=your-redis-url
UPSTASH_REDIS_REST_TOKEN=your-redis-token
```
### 2. Run locally
```bash
npx vercel dev
```
### 3. Deploy to Vercel
```bash
vercel
```
Set the same environment variables in your Vercel project settings.
### 4. Point MCP to your API
The MCP server uses `https://memo-upstash.vercel.app` by default. To use your own API, modify `index.ts`:
```typescript
const GET_URL = "https://your-api.vercel.app/api/get";
const SET_URL = "https://your-api.vercel.app/api/set";
```
## Options
### `--ttl-mins`
Set expiration time in minutes. Default is 1440 (24 hours).
```json
{
"mcpServers": {
"memo": {
"command": "npx",
"args": ["-y", "@upstash/memo", "--ttl-mins", "4320"]
}
}
}
```
## License
MIT
