Your AI agent
has amnesia.

mnemon-mcp gives it a structured, persistent memory — local‑first, zero‑cloud, one SQLite file.

OpenClaw · Claude Code · Cursor · Windsurf · any MCP client
View on GitHub
terminal
> Tell Claude about project architecture...
Claude: "Got it, I'll remember that your project
uses TypeScript monorepo with React frontend
and PostgreSQL."
new session
> What's our project architecture?
Claude: "I don't have any information about
your project architecture."
mnemon-mcp connected
> What's our project architecture?
Claude: "Based on my memory: TypeScript monorepo,
React frontend, PostgreSQL, deployed on
Vercel with GitHub Actions CI/CD."

Every session starts from zero.

Same context. Same mistakes. Every time.

What if it actually remembered?


Not all memories are equal

Different knowledge has different lifetimes. Mnemon organizes it into four typed layers.

📅
Episodic

Events & Sessions

What happened, when it happened. Decays over time — last week matters more than last year.

"Debugged auth issue on March 5"
💡
Semantic

Facts & Knowledge

Stable truths about your world. People, preferences, decisions. Doesn't decay.

"User prefers dark theme"
Procedural

Rules & Workflows

How things should work. Loaded at the start of every session automatically.

"Always run tests before commit"
📚
Resource

Reference Material

Book notes, documentation, guides. Accessed on demand when needed.

"React 19 migration notes"

Three commands. That's it.

Install

One package

Four production dependencies. No Docker, no cloud, no API keys.

bash
npm install -g mnemon-mcp
Connect

Add to your client

One command for OpenClaw. Four lines of JSON for everyone else.

openclaw mcp register mnemon-mcp --command="mnemon-mcp"
// ~/.claude/mcp.json { "mcpServers": { "mnemon-mcp": { "command": "mnemon-mcp" } } }
// .cursor/mcp.json { "mcpServers": { "mnemon-mcp": { "command": "mnemon-mcp" } } }
Done

It just works

Your agent now has persistent memory. The database is created automatically.

No setup.
No configuration.
No accounts.

Built for developers who
value their data

Air-gapped Zero network calls. Zero telemetry. Your memories stay on your machine.
Smart Search FTS5 with BM25 ranking, Snowball stemming, AND→OR fallback. No GPU.
Fact Versioning Old values aren't deleted — they're chained. Search returns latest, inspect shows history.
Single File One SQLite database. Backup = copy file. Delete = delete file.
KB Import Point mnemon at a Markdown folder. It splits, routes, and indexes everything.
MCP Native Works with Claude Code, Cursor, Windsurf — any MCP client.

Under the hood

📄 Files
Your Markdown knowledge base
🔄 Pipeline
Split, route, deduplicate, stem
🗃 SQLite
FTS5 index + BM25 scoring
🔌 MCP
stdio / HTTP transport
🤖 Agent
OpenClaw, Claude, Cursor…
TypeScript 5.9 better-sqlite3 Snowball stemmer WAL mode FTS5 triggers Versioned migrations

How it compares

Feature mnemon-mcp mem0 basic-memory Anthropic KG
Architecture SQLite FTS5 Cloud + Qdrant Markdown + vector JSON file
Memory structure 4 typed layers Flat Flat Graph
Fact versioning Superseding chains ~ Partial
Stemming EN + RU EN only EN only None
OpenClaw support Native MCP
Cloud required No Yes Optional No
Cost Free forever $19–249/mo Free + SaaS Free
Setup npm install Docker + API keys pip + deps Built-in
0
Tests
0
Dependencies
0
Memory Layers
0
Network Calls
MIT
License
🦞

Works natively with OpenClaw

OpenClaw has native MCP support. One command gives your OpenClaw agents persistent, structured memory across all sessions.

bash
openclaw mcp register mnemon-mcp --command="mnemon-mcp"

Who it's for

💻

The Solo Developer

You have coding conventions, project history, and decisions scattered across sessions. Mnemon remembers them all. Your agent starts every session knowing your stack, your preferences, and what you did yesterday.

📖

The Knowledge Worker

You keep a personal knowledge base — journal, book notes, people, projects. Mnemon imports it all and makes it searchable by your AI agent. Ask "what did I decide about X?" and get an answer.

👥

The Team Lead

Onboarding context, architectural decisions, coding standards. Store them as procedural memories and every team member's agent starts with the same foundation.


Give your agent
a memory.