Skip to content
Recall.
Get started
DocumentationGet your API key

Start here

Getting started

Connect your client, choose a workspace, and give your next session something to build on.

Using Claude Code?

The plugin brings your tools and hooks together.
Install the plugin

In this guide

#

What is Recall?

Persistent memory for your AI coding assistant

Recall gives Claude (and other AI assistants) long-term memory that persists across sessions. Instead of re-explaining your project every time, Claude remembers your decisions, coding patterns, and preferences automatically.

Without Recall

  • - Re-explain project context every session
  • - Repeat decisions you made last week
  • - Waste tokens on background info

With Recall

  • - Context loaded automatically
  • - Past decisions recalled from memory
  • - ~500 tokens vs 2000+ re-explained
1

Get Your API Key

You already have one — just copy it

Create a Recall account or sign in, then open the API Keys page in your dashboard and copy your key. It looks like sk-abc123...

Keep it secret. Your API key gives access to your stored memories. Never share it publicly or commit it to git.

2

Install Recall

Choose the plugin (recommended) or manual setup

Option A: Claude Code PluginRecommended

The Recall plugin bundles the MCP server, auto-memory hooks, RLM agents, and a setup wizard — all in one install.

1. Add the marketplace and install:

/plugin marketplace add joseairosa/recall-claude-plugin
/plugin install recall@recall-claude-plugin

2. Configure your API key:

/recall:setup

This prompts for your API key (get it from the API Keys page in this dashboard) and writes it to all the right config files automatically.

3. Restart Claude Code — the plugin MCP server reads the env var at launch.

MCP Server: Auto-configured connection to recallmcp.com
Auto-Memory Hooks: Context loads at session start, actions captured automatically
Setup Wizard: Run /recall:setup to verify your config
Auto-Updates: Plugin updates when you restart Claude Code
Option B: Manual MCP Config

Add Recall directly to your MCP config. Open or create .mcp.json in your project root (Claude Code) or ~/.claude/claude_desktop_config.json (Claude Desktop):

{ "mcpServers": { "recall": { "type": "url", "url": "https://recallmcp.com/mcp", "headers": { "Authorization": "Bearer YOUR_API_KEY" } } } }

Replace YOUR_API_KEY with the key from Step 1. Restart Claude Code to load.

3

Your First Session

Context loads automatically with the plugin — or prompt Claude manually

Plugin users: If you installed the plugin in Step 2, your context loads automatically at every session start — no prompt needed. Skip to Step 4.

Manual MCP users: at the start of a coding session, ask Claude something like:

"Set the Recall workspace to this project and load any previous context."

Claude will automatically call set_workspace and auto_session_start to load your past decisions, patterns, and preferences for this project.

What happens behind the scenes:

a

Workspace is set

Memories are isolated per project, so you only get relevant context.

b

Past context is loaded

Recent decisions, active directives, code patterns, pending to-dos, and critical items are retrieved.

c

Claude is ready

Claude now knows what happened in previous sessions and can continue where you left off.

4

Saving Memories

Tell Claude what to remember

During your session, you can ask Claude to remember anything important:

"Remember that we decided to use ULIDs instead of auto-incrementing IDs."
"Store a note that the API follows RESTful conventions with /api/v1/ prefix."
"At the end of this session, summarize what we did and save it to Recall."

Claude will use the appropriate Recall tools automatically. You don't need to know the tool names.

5

Go Fully AutomaticRecommended

Install hooks once — never do Steps 3 & 4 manually again

Auto-Memory Hooks are lightweight bash scripts that plug into Claude Code's lifecycle. Once installed, context loads automatically at session start. Failing Bash commands are captured with an output excerpt, and a session summary is stored at session end.

Plugin users (recommended)

If you installed via the plugin in Step 2, hooks are already included. No extra setup needed — they activate automatically.

Manual MCP users

curl -fsSL https://recallmcp.com/install-hooks | bash
▶

Session Start

Fetches your recent memories and injects them as context — no prompt needed.

✎

After a Bash failure

Stores a failing command and its output excerpt so the next session can learn from it.

⧖

Before Compaction

Saves a state marker so context handoff is smooth when the window fills.

■

Session End

Stores a summary of session activity when memories were recorded.

What Makes a Good Memory?

Worth Storing

  • Decisions: "Chose Redis over Postgres because..."
  • Rules: "Always use ULIDs, never auto-increment"
  • Patterns: "API endpoints follow /api/v1/{resource}"
  • Preferences: "Use informal tone, call me by first name"
  • Session summaries: "Today we built the auth system using JWT..."

Skip These

  • -Code implementations (that's what files are for)
  • -Temporary debugging notes
  • -Generic knowledge Claude already knows
  • -Casual conversation

Keep up with what's changing.

Read the changelog