Back to module
Step 1-8 min

Set Up Your Workspace

Install your AI coding tool, then have it install Bruin CLI and wire up MCP for you. Every step from here on is a plain-English prompt.

What you'll do

  1. Install an AI coding tool (Claude Code, Cursor, or Codex) - the only thing you install yourself
  2. Have the agent install Bruin CLI on your laptop for you
  3. Wire the agent to Bruin via MCP so it can run Bruin commands on your behalf

After this step you stop running terminal commands. You prompt the agent in plain English and it does the running.

Why this step matters

Every step after this is a plain-English prompt - "pull the last 90 days from Google Ads", "build a staging view", "what's our CAC by channel?". The AI agent is the one typing commands, editing files, and running SQL. You're the analyst telling it what you want. This step installs the agent and gives it the keys to the workshop.

1. Install your AI coding tool

Pick one - any of these works. Claude Code is the fastest to get running.

Claude Code

Follow the Claude Code installation guide. On Mac it's one command:

brew install claude

Sign in once (claude - it opens a browser) and you're ready.

2. Have the agent install Bruin CLI

Open your AI tool in any folder you'd like to use as your workspace, and prompt:

AI Prompt

Install the Bruin CLI on this machine by running curl -LsSf https://getbruin.com/install/cli | sh, then verify it worked with bruin --version. If bruin isn't on my PATH after install, tell me what to add or which terminal to open.

The agent will ask for permission to run a shell command - approve it. (You'll see this prompt a lot in the steps that follow; you're authorizing the agent to do the work for you.) When it finishes, the agent should print back a Bruin version number.

If your AI tool can't run terminal commands, paste the install command into any terminal yourself: curl -LsSf https://getbruin.com/install/cli | sh. The rest of the tutorial works the same.

3. Add Bruin as an MCP server

MCP (Model Context Protocol) is the bridge that lets your AI tool call Bruin commands directly. Once it's wired up, the agent can ingest, query, validate - all of it - through prompts.

Claude Code

Prompt the agent:

AI Prompt

Register Bruin as an MCP server by running claude mcp add bruin -- bruin mcp. Then remind me to restart Claude Code so the change takes effect.

Important: close your current Claude Code session and start a new one before moving on.

4. Verify the wiring

Open your AI tool and prompt:

AI Prompt

List the Bruin connections I have configured.

The agent should call Bruin, get back an empty list (you haven't added any connections yet), and report that. You want the agent to actually run the command - not just guess. If it replies with generic text like "I don't know, can you check?", MCP isn't wired. Go back and double-check the setup above.

What just happened

You have the full toolchain on your laptop: an AI brain (Claude/Cursor/Codex), a data engine (Bruin), and a bridge between them (MCP). From here on out, every step is "tell the agent what you want, the agent runs the command." You won't open a terminal yourself again unless you want to.