memento — Claude Code MCP server: local memory for AI assistants
A memory layer that lives in a local SQLite file and speaks MCP, so assistants stop asking you to re-explain your preferences and past decisions.
claude mcp add --env MEMENTO_DB=<MEMENTO_DB> --transport stdio memento -- npx -y @psraghuveer/mementoWhat memento does
Memento runs an MCP server over a SQLite database on your machine. Any client that speaks MCP can write structured notes there and read them back: preferences, project conventions, decisions you already made. By default it makes no outbound network calls, so the store stays local. You need Node.js 22.11 or newer, since the server launches through npx.
The setup starts with npx @psraghuveer/memento init. On a TTY it creates the database in the XDG location, applies migrations and asks four single-keystroke questions: your preferred name, whether to copy the bundled skill into ~/.claude/skills/, whether to seed a starter pack, and whether to write the persona snippet into custom-instructions files of clients it detects. That last question carries weight. The MCP spec treats the instructions field as optional, and a skill fires only on matching intent, so the persona block is what reliably reaches the assistant's system prompt. UI-only clients get paste-in text instead. Add --no-prompt to skip the interactive part in CI.
Starter packs fill an otherwise empty store in one command, for instance memento pack install engineering-simplicity. Four ship with the package: engineering-simplicity, pragmatic-programmer, twelve-factor-app and google-sre. You can also build a pack from memories you already have and hand it over as a file or an HTTPS link.
What it helps with
- keep preferences and project conventions in one store instead of scattered rule files
- bring last week's decisions back into a fresh chat without re-explaining
- seed an empty memory store with a ready-made engineering pack
- confirm the assistant really calls the memory server
- share your own memory collection as a file or link
Example prompts for the agent
- Remember that I use pnpm instead of npm for Node projects.
- Which package manager do I prefer?
- What did we settle on for the dashboard revamp last week?
How to install memento
Platform CLI
claude mcp add --env MEMENTO_DB=<MEMENTO_DB> --transport stdio memento -- npx -y @psraghuveer/mementoReplace <…> with your own values
Platform CLI
codex mcp add memento --env MEMENTO_DB=<MEMENTO_DB> -- npx -y @psraghuveer/mementoReplace <…> with your own values
Config file
{
"mcpServers": {
"memento": {
"command": "npx",
"args": [
"-y",
"@psraghuveer/memento"
],
"env": {
"MEMENTO_DB": "<MEMENTO_DB>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"memento": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@psraghuveer/memento"
],
"env": {
"MEMENTO_DB": "<MEMENTO_DB>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e MEMENTO_DB=<MEMENTO_DB> memento npx -y @psraghuveer/mementoReplace <…> with your own values
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about memento
How do I install memento in Claude Code?
Run claude mcp add memento -- npx -y @psraghuveer/memento, then restart the client. To point at a specific database, use the --env MEMENTO_DB=<MEMENTO_DB> variant with the stdio transport.
What does memento do?
It serves a SQLite-backed memory store over MCP. The assistant writes facts about you and your decisions into it and searches them later, so a new session does not start from zero.
Does memento need a network connection?
No. Outbound network calls are off by default. Node.js 22.11 or newer is required because the server runs via npx.
