You just finished a 2-hour debugging session with Claude Code.
You fixed the auth bug. You mapped the whole codebase together. You finally got it.
Tomorrow morning, you open a new session.
Claude has no idea who you are. No idea what project you're in. No idea you already fixed that bug. It re-reads the same files. It lands on the same wrong assumptions. You spend the first 10 minutes re-explaining everything.
This happens to everyone using Claude Code. Every. Single. Day.
If you use Claude Code more than twice a week, you must install claude-mem now.
This guide is the full playbook. How to install it. How it works under the hood. How to tune it so it doesn't eat your context window. And where it breaks (I'll be honest about the sharp edges).
Two things before we start:
Save this guide and block 20 minutes this week to install it.
Send it to anyone you've watched re-explain the same project to Claude three times.
Forget CLAUDE.md. It's not memory.
Most people think Claude Code already has memory because of CLAUDE.md files.
It doesn't.
Here's what Claude Code actually ships with, and why none of it is real memory:
CLAUDE.md → A static markdown file loaded at session start. Capped around 200 lines before Claude stops following it reliably. No search. No retrieval. You write rules once and hope they stick.
Auto Memory (v2.1.59) → Claude decides what to save. Stored as unstructured notes. No search. No filter. What gets saved rarely matches what you wanted saved.
/compact→ Summarizes your chat to free up context. Everything conversational disappears.
None of these capture what happens during a session automatically. None of them survive a crash mid-refactor. None of them let you search across projects.
claude-mem is the plugin that fills the gap. It:
Hooks into every Claude Code session lifecycle event
Captures what happens after every tool call (not at session end)
Compresses raw output into structured observations using AI
Stores everything locally in a SQLite database
Injects the relevant pieces back when you start a new session
It runs as a plugin, not an MCP server. That distinction is the whole game.
Plugins fire automatically on every session event. MCP servers sit idle until Claude decides to call them. With an MCP-based memory tool, retrieval only happens when Claude thinks to ask for it. claude-mem captures and injects without Claude having to choose to.
Install it in 2 commands (and 1 trap to avoid).
Want to keep reading?
Become a paying subscriber to get access to this post and lots of other premium content.