Every time I start a new AI coding session, I have to explain everything from scratch.
"We use PostgreSQL because..."
"Our auth uses JWT with 24h expiry..."
"The API follows REST with /api/v1/..."
The AI forgets. Every. Single. Time.
After the 50th time explaining the same decisions, I built LORE.
What is LORE?
LORE is an MCP server that reads your existing codebase and automatically extracts architectural decisions — no manual work needed.
cd your-project
lore init
That's it. LORE generates a LORE.md with everything your AI needs to know:
## Database
### 🟢 PostgreSQL as primary database
**Why**: Found pg/postgres in dependencies
**Rules**: Use connection pooling always
## Authentication
### 🟢 JWT-based authentication
**Rules**: Never store secrets in code
The Problem It Solves
AI coding tools like Claude Code and Cursor are powerful. But they have zero memory between sessions.
Every new session = complete amnesia.
Your team spent months deciding:
- Why PostgreSQL over MongoDB
- Why JWT over sessions
- Why Zod over Yup
The next AI session knows none of this.
LORE remembers so you don't have to repeat yourself.
How It Works
LORE reads your:
-
package.json→ detects frameworks, libraries, databases -
.env.example→ detects security patterns -
docker-compose.yml→ detects deployment decisions
Then generates a structured LORE.md that plugs directly into your MCP settings.
Quick Start
npm install -g lore-mcp
cd your-project
lore init
lore status
MCP Integration (Claude Code / Cursor)
{
"mcpServers": {
"lore": {
"command": "node",
"args": ["/path/to/lore-mcp/dist/index.js"]
}
}
}
What's Next
-
lore doctor— diagnose setup issues ✅ -
lore status— view all decisions ✅ - LORE INTEGRITY — verify decisions are actually implemented in code
- VS Code Extension
- LORE NETWORK — share anonymous architectural patterns
Open Source
LORE is fully open source and local-first. No data leaves your machine.
⭐ github.com/EliotShift/lore-mcp
Built in one day from Morocco 🇲🇦
Top comments (0)