Home/News/How to Add Persistent Memory to Claude Code with the Endpoints Memory Plugin

How to Add Persistent Memory to Claude Code with the Endpoints Memory Plugin

Adam Carter
How to Add Persistent Memory to Claude Code with the Endpoints Memory Plugin

The Problem

Claude Code is incredibly powerful for software engineering, but it has one frustrating limitation: it forgets everything between sessions. Every time you start a new conversation, you're back to square one.

That research you did yesterday about your codebase architecture? Gone. The decision you made about which library to use? Forgotten. The context about your project's quirks and conventions? You'll have to explain it all over again.

I got tired of repeating myself.

The Solution

I built the endpoints-memory plugin to give Claude Code a persistent memory system. It uses a two-tier architecture:

  • cc-ram: A quick-lookup index of everything Claude remembers
  • cc-drive: Deep storage for full research findings, decisions, and context

When you start a new session, Claude automatically loads the memory index and can recall what it knows. After doing research (web searches, codebase exploration), it prompts to save the findings. Before context gets compacted, it reminds you to preserve important learnings.

How I Use It

Here's my typical workflow:

Session Start: The plugin loads cc-ram automatically. Claude confirms what it remembers: "I have 7 memories available" and lists topics like plugin architecture, API patterns, and project conventions.

During Research: When I ask Claude to search the web or explore the codebase, a hook fires after the results come back, prompting Claude to save the findings to cc-drive.

Before Compaction: When context is about to be summarized (losing detail), the PreCompact hook reminds me to save anything important.

For example, I've saved research about Claude Cowork's plugin system, Claude Code's transcript storage locations, and security considerations for plugin development. Next session, I don't have to re-research any of it.

Step-by-Step Setup

Step 1: Sign Up for Endpoints (Free)

Go to endpoints.work and create a free account. You'll use Endpoints as the backend storage for your memories.

Endpoints signup page

Step 2: Get Your API Key

After signing up, navigate to your dashboard and copy your API key. It starts with ep_.

API key in dashboard

Step 3: Add the Plugin Marketplace

In your terminal, add my plugin marketplace to Claude Code:

``bash claude plugins marketplace add adamkristopher-plugins ``

Adding the marketplace

Step 4: Install the Plugin

Now install the endpoints-memory plugin:

``bash claude plugins install endpoints-memory ``

Step 5: Login

Start a Claude Code session and run the login command:

`` /login ``

When prompted, paste your API key (the one starting with ep_).

Login success

Step 6: Verify It Works

Run the recall command to check your memory:

`` /endpoints-memory:recall ``

You should see "No memories yet" if this is your first time, or a list of stored memories if you've already saved some.

Recall command output

Using the Plugin

The plugin provides three main commands:

CommandWhat It Does
/endpoints-memory:recallList all stored memories from cc-ram index
/endpoints-memory:saveSave the current session's learnings
/endpoints-memory:endpointsDirect access to the Endpoints API

You don't need to remember to save manually most of the time. The plugin includes hooks that:

  1. SessionStart: Loads your memories when you begin
  2. PostToolUse: Prompts to save after web searches or exploration
  3. PreCompact: Reminds you to save before context compression

How It Works Under the Hood

Memory architecture diagram

The plugin uses Endpoints as a structured storage backend:

cc-ram/index: A single endpoint that holds an index of all memories. Each entry has a slug and summary for quick lookup.

cc-drive/{slug}: Individual endpoints for each topic. Contains full research findings, sources, key discoveries, and context.

  • The Stack:
  • 3 hooks (SessionStart, PostToolUse, PreCompact)
  • 3 skills (config, endpoints, memory commands)
  • 1 sub-agent (memory-saver)

When you save, the memory-saver agent extracts key findings from the conversation and stores them in a structured format that's easy to retrieve later.

What's Next

This plugin is also compatible with Claude Cowork since both products share the same plugin architecture. I'm working on more specialized plugins that use the same Endpoints backend:

  • Leads Vault: Save and track sales leads
  • Receipt Keeper: Scan and categorize expenses
  • Source Library: Build a research bibliography

The plugin is open source. Check it out on GitHub or install it from the marketplace to start building your Claude's long-term memory.


Questions or feedback? Reach out at [endpoints.work/contact](https://endpoints.work/contact).