mcp server

mnemoir

Offline-first persistent memory system for AI coding agents

alle-bartoli/mnemoircommunitylistedunreviewed

Use tool

Set up

Runs on your computer. Pick your agent and operating system.

This project does not publish an install method we can turn into a config. Follow the README instead.

README

synced Oct 26, 2026

Mnemoir

MCP server that gives AI coding agents long-term memory. Mnemoir runs over stdio and stores memories in a local Redis Stack instance.

Offline-first: local ONNX embeddings and rule-based compression; no API keys required
Hybrid search: semantic, keyword, and importance scoring
Session-aware: automatic summaries and spaced-repetition recall
Multi-project: memories are scoped by project

Quick start

Requirements

Go 1.25+
Docker with Compose v2
Task 3.x
jq (used by setup hooks)
Git; Windows users need Git for Windows

Create a local .env with a strong Redis password, then run the setup for your client:

git clone https://github.com/alle-bartoli/mnemoir.git
cd mnemoir
echo 'MNEMOIR_REDIS_PASSWORD=your-secret' > .env
task setup

task setup starts Redis, builds Mnemoir, copies the default configuration, prewarms the local embedding model, and configures Claude Code when it is installed. For Codex CLI use task setup:codex.

For manual clients, use ~/.mnemoir/.env instead of a repository .env when the client starts outside the repository directory.

The default configuration is fully local and does not require API keys. Redis still runs locally through Docker.

Supported clients

Client	Setup	Session hook
Claude Code	`task setup`	`SessionEnd`
OpenAI Codex CLI	`task setup:codex`	`Stop`
Claude Desktop, Pi, Cursor, Windsurf, Cline, Zed	manual setup	Manual `end_session`

All MCP-compatible clients can use Mnemoir. Clients without a session hook should call end_session before the conversation ends.

Configuration

The setup copies config/default.toml to ~/.mnemoir/config.toml. Edit that file to change providers, model paths, Redis, scoring, or the sideband health server.

The default providers are:

Provider	Default	API key
Compressor	`local`	None
Embeddings	`local`	None

See client setup for manual configuration and optional provider keys.

Common commands

task setup             # Install for Claude Code
task setup:codex       # Install for Codex CLI
task prewarm           # Download the local embedding model
task test              # Run tests
task build:all         # Cross-compile supported platforms
task docker:up         # Start Redis Stack
task docker:down       # Stop Redis Stack
task backup            # Native Redis backup
task backup:json       # Portable JSON backup
task uninstall         # Remove installed artifacts

Documentation

Client setup: manual MCP configuration and platform notes
Troubleshooting: common startup and installation problems
Backup and restore: native and JSON backups
Search scoring and retention: hybrid ranking, decay, and auto-forget
Agent specs: instructions for AI agents using Mnemoir

Development

task --list    # List all tasks
task test      # Run the test suite
task build     # Build for the current platform

License

MIT