Building Your First Hex: The 90-Minute Setup

You've read the framework. You understand the six slots. You get why constraints work. Now the question is whether you'll actually build the thing or whether this becomes another PDF in your Downloads folder next to the other frameworks that were going to change everything. This article is the 90-minute walkthrough — from "I have a Claude subscription" to "I have a working hex that does something useful." Ninety minutes is aggressive. It's also realistic, provided you don't do the thing that kills most setups before they start: overthinking the tool selection.

The Prerequisite

You need a Claude Pro or Team subscription with MCP access. That's the foundation. Everything in the hex connects through Claude's MCP architecture, so if you don't have MCP access, you don't have the infrastructure. As of early 2026, MCP is available on Claude Pro ($20/month), Team ($25/month per seat), and Enterprise plans. The free tier does not include MCP. If you're on the free tier, this is the one subscription that matters — upgrade before buying anything else [VERIFY — current Claude subscription tiers and MCP access as of early 2026].

You also need a computer where you can run terminal commands. MCP connectors install and run through the command line. If the word "terminal" makes you nervous, take a breath — the actual commands are straightforward, and I'll walk through them. But if you've never opened Terminal on a Mac or Command Prompt on Windows, budget an extra thirty minutes for getting comfortable with the interface.

Everything else — the specific tools for your six slots, the accounts, the API keys — you probably already have. The hex isn't asking you to buy new things. It's asking you to connect what you already own.

Step 1: Choose Your Six (15 Minutes)

Open the slot framework from the previous article. Six slots: thinking, publishing, communication, automation, data/knowledge, creation. For each slot, write down the tool you already use. Not the tool you want to use. Not the tool that impressed you in a demo. The tool you actually opened this week.

If you don't have a tool for a slot, leave it empty. A hex with four active skills and two empty slots is a legitimate hex. It's honest about your current work and leaves room to grow. A hex with six tools where two are aspirational is a wish list, not a system.

Here's what a typical first hex looks like for a content creator:

• Slot 1 (Thinking): Claude
• Slot 2 (Publishing): Ghost
• Slot 3 (Communication): Gmail
• Slot 4 (Automation): empty — no automations running yet
• Slot 5 (Data): local filesystem
• Slot 6 (Creation): GPT Image Generation

And here's what it looks like for someone in operations or consulting:

• Slot 1 (Thinking): Claude
• Slot 2 (Publishing): Google Docs (client deliverables)
• Slot 3 (Communication): Gmail
• Slot 4 (Automation): n8n
• Slot 5 (Data): Google Drive
• Slot 6 (Creation): empty — doesn't produce media

Both are valid hexes. The framework is the same. The contents reflect the actual work.

Spend fifteen minutes on this. Set a timer. The most common failure mode at this step is spending two hours agonizing over whether Notion or Obsidian is the better knowledge tool. It doesn't matter. Pick one. You can swap in three months. The cost of picking the "wrong" tool for a slot is near zero. The cost of spending three hours deciding is three hours you didn't spend building.

Step 2: Install MCP Connectors (30 Minutes)

Each tool in your hex needs an MCP connector — the bridge that lets Claude access it. Here's what the installation process actually looks like.

On a Mac, you'll open Terminal. On Windows, you'll use Command Prompt or PowerShell. The Claude desktop app has a configuration file — typically at ~/Library/Application Support/Claude/claude_desktop_config.json on Mac — where you specify which MCP servers to run. Each connector is defined as a server entry in that file with a command to start it and any required environment variables (API keys, credentials).

For filesystem access, the connector is built into Claude's architecture. You point it at a directory, and Claude can read and write files there. This is the easiest connector to set up and the most universally useful. If this is the only one that works in your first session, you still have a functional hex.

For Google services — Gmail, Calendar, Drive — the setup involves authenticating with your Google account through OAuth. The MCP connectors for Google services are well-maintained and the authentication flow, while slightly involved, is a one-time process. You'll be redirected to a Google consent screen, grant permissions, and the connector stores the token locally. Community-maintained guides walk through this step by step [VERIFY — check that current Google MCP connector setup guides are accurate for the March 2026 versions].

For other tools, the process varies. Ghost has an MCP connector that requires your Ghost Admin API key and site URL. n8n can be triggered through webhooks that Claude calls via MCP. Image generation through GPT's API requires an OpenAI API key. Each connector has its own setup, but they all follow the same pattern: install the connector, add credentials, register it in Claude's config file.

The thirty-minute budget assumes you're setting up three to four connectors. If you're setting up all six, give yourself forty-five minutes. If one connector is giving you trouble — authentication failing, server not starting — skip it. Come back to it later. The hex works with partial connectivity. Getting stuck on one connector and burning your entire session troubleshooting is the second most common failure mode after overthinking tool selection.

Step 3: Test Each Connection (15 Minutes)

For each connected tool, give Claude a simple task that proves the connection works.

Filesystem: "List the files in my Documents folder." If Claude returns a file list, the connection works.

Gmail: "Show me my three most recent unread emails." If Claude returns email subjects and senders, the connection works.

Calendar: "What's on my calendar tomorrow?" If Claude returns your actual events, the connection works.

Publishing tool: "Create a draft post titled 'Test Post' in Ghost." If the draft appears in your Ghost admin, the connection works.

This is a diagnostic pass, not a production test. You're checking that the pipe is connected, not that the water pressure is perfect. Each test should take one to two minutes. If a test fails, note it and move on. A hex with four working connections and two broken ones is still more useful than no hex at all. Fix the broken ones tomorrow.

Step 4: Build Your First Cross-Tool Workflow (15 Minutes)

This is where the hex becomes a system instead of a list. Give Claude a task that requires at least two of your connected tools to work together.

The simplest cross-tool workflow: "Check my calendar for today and draft an email to [name] summarizing what meetings I have." This uses your calendar connector and your email connector. Claude reads the calendar, composes the email, and either sends it or presents it for your approval.

A content-focused version: "Read the markdown file at /content/drafts/new-article.md, generate a featured image that matches the topic, and create a draft post in Ghost with the content and image attached." This uses filesystem, image generation, and publishing — three tools in one request.

An operations version: "Search my email for messages from [client name] this week, summarize the open action items, and create a Google Doc called '[Client] Action Items — [Date].'" This uses email, data processing (Claude's native capability), and Google Drive.

If the cross-tool workflow works — even partially, even with a hiccup — your hex is functional. The tools are connected and Claude can orchestrate between them. That's the milestone. It doesn't have to be elegant. It has to work once. You'll refine it over the coming weeks.

Step 5: Write Your System Prompt (15 Minutes)

The system prompt is the two-hundred-word instruction set that tells Claude how you work. It's not a personality description. It's not a life story. It's operational context — what your tools are, what your workflow looks like, and what kinds of tasks you'll be asking for.

A functional system prompt looks like this:

"I'm a freelance content creator. I publish articles on Ghost, manage communication through Gmail, and store all project files in /Users/me/Projects/. When I ask you to publish something, always create it as a draft first — never publish directly. When I ask about my schedule, check Google Calendar. When I ask for images, generate them at 1200x630 for featured images unless I specify otherwise. My writing style is direct and informal. Don't add emojis to anything."

That's it. That's a working system prompt. It tells Claude what tools are available, what the defaults should be, and what your preferences are. You can refine it over time as you notice patterns — "always use these tags when publishing to Ghost," "never schedule meetings before 10am," "when drafting client emails, always use professional tone." The system prompt grows with usage. The initial version just needs to be functional, not complete.

Where People Get Stuck

Three failure modes dominate first hex builds.

The first is connector installation anxiety. The terminal commands look intimidating if you've never used a command line. The JSON config file looks fragile. The OAuth authentication flows feel like you're one wrong click from breaking something. The reality is that MCP connector installation is a solved problem with good documentation — but the feeling of unfamiliarity is real, and it causes people to close the terminal and "come back to this on the weekend." If this is you, commit to the filesystem connector only. That one requires no external authentication and works with minimal config. Get one connector running. The confidence from that carries you through the others.

The second is tool selection paralysis. You spent two hours reading comparison articles about Notion versus Obsidian instead of picking one and moving on. The hex framework specifically tells you that the tool choice is less important than the commitment — but knowing that doesn't always override the instinct to optimize. If you're paralyzed, default to the tool you've used most recently. That's the heuristic. Don't research. Don't compare. Just pick the one you opened last week.

The third is system prompt perfectionism. You want the system prompt to anticipate every possible use case before you start using the hex. It can't. The first version will be incomplete. You'll use the hex for a week and realize you forgot to mention how you want your calendar events formatted. Good. Update the prompt. The system prompt is a living document, not a spec sheet. Write it in fifteen minutes and revise it after your first real work week.

What "Working" Looks Like

At the end of ninety minutes, you should have: Claude with MCP access to at least three of your six tools, one successful cross-tool workflow (even if it was simple), and a system prompt that gives Claude basic operational context.

That's a working hex. It's not a finished hex — it's going to evolve over weeks and months as you use it, swap tools, refine prompts, and discover workflows you didn't anticipate. But it's working, which puts you ahead of the overwhelming majority of people who downloaded the PDF, admired the framework, and never built anything. The difference between a hex on paper and a hex that runs is about ninety minutes and the willingness to pick tools without agonizing. You've done the hard part. Now use it.

This is part of CustomClanker's Hex Explained series — you downloaded the hex, here's context.