Docs
Documentation
DocumentationGetting Started

Getting Started

Connect Recordari to your AI agent in five minutes.

Recordari gives your AI agents a memory that persists across sessions. Instead of starting every conversation from scratch, your agent loads its memory at the start of a session and picks up exactly where it left off - with context about past decisions, active constraints, and recent work.

This page gets you from zero to a working connection in five minutes.

What you need

  • A Recordari account - sign up at recordar.io
  • An MCP-compatible AI agent: Claude Code, Claude.ai, or any client that supports the Model Context Protocol

Step 1 - Get an API key

After signing up, open the admin portal and go to API Keys → Personal keys. Click New personal key, give it a name (e.g. "Claude Code - my-project"), and copy the value immediately. You won't be able to see it again.

A personal key is tied to you individually - it's what you want for your own agent sessions. See Authentication for how this differs from a shared workspace key.

Step 2 - Install the CLI and run the setup wizard

The fastest path is to install the recordari CLI and run recordari init. The wizard handles everything in one pass: confirms your API key, installs the agent skill, wires your CLAUDE.md, and configures lifecycle hooks for your platform.

# macOS / Linux
curl -fsSL https://pub-21464170f08f4747b42f28910795a7ad.r2.dev/install.sh | sh
# Windows (PowerShell)
irm https://pub-21464170f08f4747b42f28910795a7ad.r2.dev/install.ps1 | iex

Once installed:

recordari config set api_key YOUR_API_KEY
recordari init

The wizard detects Claude Code, Cursor, VS Code Copilot, and Codex automatically. Run recordari doctor to confirm everything is working. See Recordari CLI for the full command reference.

Manual setup — Claude Code

If you prefer to skip the CLI, create .mcp.json at the root of your project:

{
  "mcpServers": {
    "recordari": {
      "type": "http",
      "url": "https://api.recordar.io/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Add .mcp.json to your .gitignore before committing. Then reload Claude Code — Recordari should appear in your active MCP servers.

Manual setup — Claude.ai

Go to Settings → Integrations → Add integration, and enter:

  • URL: https://api.recordar.io/mcp

Claude.ai connects via OAuth 2.1 - you'll be prompted to log in and authorize the connection, no API key needed.

⚠️ Note: At present, using OAuth connectors with Claude.ai and ChatGPT requires the agent to be in developer mode. Check your own agent's docs to find out how to enable this if the connection doesn't work.

Step 3 - Start your first session

Tell your agent to orient on a domain. A domain is just a named bucket for a set of memories - use your project name:

"Before we start, check what you know about the backend-api domain and tell me where things stand."

On the first call the domain is empty, so you'll get a minimal response. That's fine - we're about to fill it.

Step 4 - File your first memory

Now tell your agent to remember something:

"File a memory that we're using PostgreSQL as the main database."

Your agent will file the memory and return an ID for it. If the reasoning isn't obvious from what you said, it may ask — you don't need to spell out "why it matters" every time.

Step 5 - Start the next session

Open a completely new conversation. Tell your agent:

"Before we start, check what you know about the my-project domain and tell me where things stand."

This time it will come back with your PostgreSQL decision in context. That's it - persistent memory is working.

Admin portal

Day-to-day workspace management (keys, usage, audit log) happens at admin.recordar.io, not through your agent. See the Admin portal overview for what is available on your plan.

Where to go next

Docs
Copyright © Binary Soup Software Ltd. All rights reserved.