Claude Code for GTM: Where to Start

I remember looking at a tool called Clay back in 2022.

At the time, many just saw it as another outbound tool. It felt technical, and frankly, a little intimidating too.

But I saw something else. I saw a tool that was going to fundamentally break the way we build lead lists.

If you listened to me then, you know what happened next. Clay became the operating system for modern GTM.

I have that exact same feeling today.

We are standing at another turning point, and this time it was launched by a single command in a terminal window: Claude Code.

This is not AI hype. It’s a structural shift in how companies are built.

For the last decade, we have been living in the No-Code era. We paid premiums for visual interfaces, drag-and-drop builders, and colourful buttons because they made us feel safe.

We treated the Terminal - that black screen with the white text - as a “Do Not Enter” zone reserved for engineers.

Claude just gave the rest of us keys to the engine room. And if you are willing to get over the initial fear of the black screen, you are about to realise that you don’t need to hire a developer to build your GTM stack.

You just need to learn how to ask the computer to do it for you.

We are going to cover the first steps in getting to work with it:

• What is Claude Code (and why the terminal matters).

• How to install it (even if you have never written a line of code).

• How to run your first agent to build a lead list on autopilot.

What is Claude Code?

People often think using AI means opening a browser tab to chat with a bot. You paste some text, it gives you some text back, and then you have to go do the actual work yourself.

Claude Code is different because it acts as an Agent that lives in your computer’s terminal.

The standard chat interface is passive, but Claude Code is active. When you give it permission, it actually creates files on your desktop, runs scripts, and searches the web. If it runs into an error message, it reads it and fixes its own code without you even needing to intervene.

Think of it like a very fast junior developer who has access to your keyboard. If you tell a standard chatbot to find the pricing page for 50 competitors, it might give you a Python script that you don’t know how to use.

If you tell Claude Code to do the same thing, it just gets to work. It writes the script, runs it, handles the anti-scraping blockers, and saves a CSV file called competitors.csv right to your desktop.

You didn’t write a single line of code, but the work got done.

The Terminal

When a non-technical founder sees a black terminal window, they usually panic. It looks like Matrix code where one wrong keystroke deletes the company database.

But using the terminal just means cutting out the middleman. Here is the entire setup process to get started. It takes about 5 minutes.

The Setup (One Time Only)

Step 1: Install Node.js: Claude Code runs on Node.js. Go to nodejs.org, download the LTS version, and run the installer. Click through the defaults. When it’s done, you can verify it worked by opening any terminal and typing node --version. If you see a number, you’re good.

Step 2: Get the Editor: Download and install VS Code (Visual Studio Code). This is where you’ll manage your files and access the terminal.

Step 3: Open the Command Center: Open VS Code. In the top menu bar, click Terminal > New Terminal. A small window will pop up at the bottom of your screen. That is your cockpit.

Step 4: Install the Brain Copy and paste this exact command into that terminal window and hit Enter:

npm install -g @anthropic-ai/claude-code

It will churn for a moment while it downloads. When it finishes, you’re done. You’ve installed the agent.

You’ll also need an Anthropic API key (pay-as-you-go, costs pennies per task) or a Claude Pro/Max subscription. Claude Code will walk you through connecting your account the first time you run it.

From now on, whenever you want to start work, you just open this terminal and type claude.

Building a Lead List

Let’s look at a concrete example. If you wanted to build a list of the latest Y Combinator startups in the AI sector, you have two bad options: pay for a scraping tool or hire a Virtual Assistant to copy-paste for three hours.

With Claude Code, this becomes a short conversation.

Step 1: Wake the agent. Open your terminal, navigate to your project folder, and type claude to start the session.

Step 2: Give the order. You don’t need code. Just type in plain English:

“Search for the top 20 Y Combinator startups in the Winter 2025 batch that focus on AI sales. Find the names of their founders. Save the company name, founder name, and website into a CSV file called ‘leads.csv’.”

Step 3: Watch the execution. Claude doesn’t just write the script, it runs it too. It will search for the data, parse what it finds, and write the CSV file directly into your project folder.

A note on expectations: this isn’t magic. Sometimes the source it finds is incomplete, or it needs a follow-up nudge from you (”You missed the last 5 — try searching the YC directory directly”). Think of it like managing a fast but junior intern. The first pass gets you 80% of the way there. A couple of corrections gets you to 95%.

When it finishes, a file called leads.csv appears in your folder. You didn’t write a single line of Python. You just managed the intern.

MCP: The Secret Weapon

If you only use Claude to write files on your desktop, you’re scratching the surface. The real power comes from a standard called the Model Context Protocol (MCP).

Think of a standard AI model as a brain in a jar. Smart, but no hands. It can plan a trip, but it can’t book the flight.

MCP gives the brain hands. It acts as a universal plug that lets Claude connect to real-world tools: search engines, databases, APIs.

Instead of buying an enrichment tool, you can plug Claude into Brave Search (a free search API) and have it act as your own private researcher.

How to Set It Up

Step 1: Go to brave.com/search/api and create a free account. The free tier gives you 2,000 searches per month, which is more than enough. Once you’re in the dashboard, generate an API key. Copy it.

Step 2: Open your regular terminal (not inside a Claude session). Navigate to your project folder and run this command, replacing YOUR_KEY_HERE with the key you just copied:

claude mcp add brave-search -e BRAVE_API_KEY=YOUR_KEY_HERE -- npx -y @modelcontextprotocol/server-brave-search

That’s one command. It tells Claude Code to connect to the Brave Search server and use your API key to authenticate.

Step 3: Start a new Claude session by typing claude. Then verify it worked:

“What MCP tools do you have available?”

If you see brave_web_search in the list, you’re live. Claude can now search the internet in real time.

What This Unlocks

Now you can run prompts like this:

“For every company in my ‘leads.csv’ file, search the web for the CEO’s name and their most recent podcast appearance or interview. Add a column called ‘Icebreaker’ to the CSV with a one-sentence reference to that appearance.”

Claude goes out to the web, researches each person individually, and updates your file. No scraping tool. No VA. Just a plain English instruction.

Context Files: The New Prompt Engineering

In the ChatGPT era, you learned “Prompt Engineering.” In the Agent era, you’ll learn Context Engineering.

You don’t paste instructions every time. Instead, you create a “Rule Book” that lives in your project folder. This file is called CLAUDE.md.

Here’s the key insight: when you type claude inside a folder that contains a CLAUDE.md file, the agent reads it automatically. It applies every rule in that file to everything it does — without you saying a word.

This file is the Constitution for your agent. It tells it who it is, what the product does, and most importantly, what it is forbidden to do.

The CLAUDE.md Template

Create a file named CLAUDE.md in your project folder and paste this in:

# GTM AGENT: [YOUR COMPANY NAME]

## 1. IDENTITY & GOAL
- **Role:** You are the Lead RevOps Engineer for [Company Name].
- **Objective:** Identify high-intent prospects in [Industry] and prepare
  technically accurate outreach data.
- **Core Value:** Precision over volume. 10 perfect leads beat 100 mediocre ones.

## 2. THE PRODUCT
- **What it is:** [1-sentence description of your product].
- **Value Prop:** "[Your main promise/benefit]."
- **Killer Feature:** We integrate directly with [key integration or unique capability].

## 3. IDEAL CUSTOMER PROFILE (ICP)
- **Company:** [Size/Revenue range] in [Geography].
- **Title:** [Job titles, e.g., CTO, VP of Sales].
- **Buying Signal:** Look for companies that have [specific trigger, e.g.,
  raised funding, posted a job for X role, installed Y software] in the last 30 days.

## 4. THE "NO-GO" LIST
- **Never target:** [Agencies / consultancies / companies under X employees].
- **Skip these industries:** [Industries that don't fit].
- **Competitors:** Never reach out to [Competitor 1], [Competitor 2].
- **Geography:** Ignore [regions you don't sell to].

Without this file, you explain your business every time you open the terminal. It’s like hiring a new intern every morning.

With CLAUDE.md, the agent has permanent memory.

It knows the Trigger: it doesn’t just look for “CTOs”. It looks for CTOs who just raised funding or posted a specific job.

It knows the Product: it understands your integrations without you spoon-feeding details.

It knows the No-Go List: it skips agencies, wrong geographies, and competitors automatically. This is the part most people miss. Telling the agent who to ignore saves you from burning your domain on bad-fit prospects.

Fill in the brackets. Save the file. Now just type “Find me 5 leads” , and Claude applies every rule without you saying a word.

The Danger Zone

I need to give you a warning before you start running commands.

Claude Code is fast. That is its biggest feature, but it is also its biggest risk. It makes mistakes faster than you can catch them.

Before, manual work was actually a safety mechanism. When you had to copy and paste email addresses one by one, you naturally slowed down. You noticed if a name looked weird or if a company didn’t fit your list. That friction gave you time to think.

Claude removes the friction entirely.

Say you tell it: “Personalise this email based on the company’s latest news.”

It scans Row 42. That company just announced a massive layoff and a lawsuit. Claude reads the headline but lacks the empathy to understand the context.

It writes: “Huge congrats on the restructuring news! Exciting times ahead.”

That draft goes into your outreach file. If you’re not paying attention and you blast it out, that email lands in the CTO’s inbox, along with 49 more just like it.

If you are reckless, you can torch your reputation in sixty seconds.

Treat this like a loaded weapon. Never let the agent fire without a safety check.

The “Safety First” Protocol

Rule #1: The Dry Run. Always ask Claude to draft, never to send. (Claude Code doesn’t send emails on its own unless you’ve specifically wired up a sending tool, but building the habit now protects you later.)

• ❌ Bad: “Email everyone in leads.csv.”

• ✅ Good: “Draft emails for the first 5 people in leads.csv. Save them to a file called drafts.txt so I can review them.”

Rule #2: The Batch Limit. Never process a whole file at once. If you have 500 leads, start with 10. If Claude hallucinates or writes a bad subject line, you want to catch it when it costs you 10 leads, not your entire market.

Rule #3: The Eye Test. Before you approve anything, open that drafts file and read the outliers. Check the companies with recent news. Did Claude mistakenly congratulate a CEO on a lawsuit? Did it misinterpret a “hiring freeze” as “growth”?

The agent does the work. You do the judgment. That’s the deal.

The Takeaway

We are witnessing the end of Software as a Service and the beginning of Service as Software.

You are no longer a User.

For the last twenty years, you paid for software that helped you do the work.

• Salesforce helped you organize data, but you had to type it in.

• ZoomInfo gave you the phone number, but you had to make the call.

• Mailchimp sent the email, but you had to write it.

Claude Code changes the contract. Instead of paying for a tool to help you, you are paying for an agent to do it for you.

This creates a new responsibility for you.

When you use HubSpot, you are a User. You click buttons. When you use Claude Code, you are a Manager. You give orders.

If you are a bad manager who gives vague instructions, your employee will produce garbage. The only difference is that Claude will produce garbage at 1,000 words per minute.

The terminal is now the most direct way to speak to your new workforce.

Jasper

---

PS: I’ve packaged an entire 4-Phase Claude Code architecture into the The CLAUDE CODE GTM Starter Guide. This transforms Claude from a simple chatbot into a Strategic Analyst.

Inside the Pack:

• 1: The Master CLAUDE.md: The 50-line “Constitution” that forces Claude to adopt your voice, ignore “Agency” leads, and think like a senior partner.

• 2: The Hands Setup (MCP): A 2-minute guide to installing Brave Search so your agent can browse the live web (LinkedIn, Crunchbase, News).

• 3: The Researcher Script: This script visits the company website, reads their Careers page to spot hiring trends, and categorises them by intent.

• 4: The Deep Dive Dossier: A script that takes your top 5 prospects and generates a 1-page briefing doc for each. It summarises their recent news, identifies their tech stack, and writes a “Why Now?” hypothesis for your pitch.

Get access here.