AI agents forget decisions
the moment a session ends.
Why you dropped an approach, why you picked this name — spec files keep the conclusion and lose the reasoning. LinkLore is project memory AI writes and reads directly. The next session's AI picks up every decision and pitfall with one brief() call.
Install — tell your agent to do it
You don't set up an MCP server by hand — your AI agent does. Paste this as your first message to Claude Code, Cursor, or any MCP-capable agent.
LinkLore(프로젝트 메모리용 MCP 서버)를 이 저장소에 설치해줘. 패키지 llre — PyPI(uvx llre)와 npm(npx llre) 둘 다로 실행돼 — 별도 설치나 클론이 필요 없어. 지금 쓰는 클라이언트에 맞게 MCP 서버로 등록해(Claude Code라면 claude mcp add llre -- uvx llre, 아니면 이 프로젝트의 .mcp.json에 {"mcpServers": {"llre": {"command": "uvx", "args": ["llre"]}}} 추가), 그다음 여기서 uvx llre init을 실행해줘. uvx llre login은 브라우저로 구글 로그인을 여는 거라 개인 백업이나 팀 공유가 필요할 때만 쓰는 거니까, 실행하기 전에 나한테 먼저 물어봐줘. 설치가 끝나면 자동으로 추가된 CLAUDE.md 섹션을 따라 — 매 세션 시작 시 brief()를 부르고, 결정은 lore로 기록해.30-second start
The real flow after install — or type it by hand with one uvx line (npx llre works the same on Node.js).
# register in .mcp.json, or just run it $ uvx llre MCP server started — stdio ───────────────────────────────────── # AI records a decision while working $ add(type='lore', title='Always confirm before push', msg='Auto-push on every commit kept failing CI — switched to always asking the user, regardless of dirty state') [lr-66189d14] Always confirm before push #git-guard ───────────────────────────────────── # next session — new chat, same project $ brief() Recent activity: [lore] always confirm before push — auto-push on every commit... [lr-66189d14]
Different chat, days later — brief() still surfaces it. No one has to copy it into CLAUDE.md by hand.
Mental model — just lore and doc
Lore
Decisions, dead ends, lessons, rules — the "why." A git commit records the what; lore records the why.
Doc
Describes the current state — feature lists, architecture, checklists. Stale detection flags it when the code moves on.
6 core tools
| Tool | Role |
|---|---|
brief() | Session start — loads open items, recent activity, key rules |
add() | Create lore/doc — decisions, specs |
show() | Search/lookup — natural language, tags, or id |
edit() | Edit — append (default), section, or overwrite |
link() | Link lore↔doc or lore↔lore |
local() | Move/view across projects on the same disk |
The full API reference (23 tools, every argument) is on the GitHub README.