My exact Claude Code setup
Claude Code is only as useful as what it can reach. This is the exact setup I run on Windows 11: one install, two CLIs, eleven connectors, and persistent memory, in the order that gets you to a working system fastest. Every connector listed here is one I actually have wired in, with what each one does for me.
## the process
- 1
Decide what Claude is allowed to touch before you install anything
Claude in an empty terminal is a chat window. Claude connected to the tools your business actually runs on is an operator. Before installing, write down the systems where your real work lives: task manager, docs, chat, email, calendar, automations, meeting notes. That list is your connector list, and nothing else gets connected. If a tool is not on the list, it does not get OAuth access, full stop.
- 2
Install Claude Code and sign in
I run the native Windows install of Claude Code on Windows 11, no WSL required for any of this setup. Install it from the official Anthropic docs, open a terminal, run claude, and it walks you through browser sign-in. I sign in with a Claude subscription rather than an API key, which means no key files sitting on disk and usage is covered by the plan. After sign-in, run claude in any project folder and confirm you get a working session before going further.
- 3
Auth the deploy path: GitHub CLI and Vercel CLI
Do this before connectors, because shipping is the whole point. Install the GitHub CLI and run gh auth login, then install the Vercel CLI with npm i -g vercel and run vercel login. Each is a one-time browser auth. From then on Claude can create repos, commit, push, open pull requests, and deploy to production without you touching a browser. Everything I have shipped with this setup goes out through these two CLIs.
- 4
Connect the work-management layer: ClickUp, Notion, Slack
I connect tools through my Claude account's connector settings, authorizing each with OAuth, and they show up inside Claude Code automatically. ClickUp means Claude can search the whole workspace, so asking for every overdue task across all lists is one prompt instead of ten filters. Notion means Claude can find and update an SOP page without me opening Notion at all. Slack means Claude can draft a channel update about what shipped and post it once I approve the wording.
- 5
Connect email, calendar, and files: Gmail, Google Calendar, Google Drive
These three are the difference between Claude knowing about your projects and Claude knowing about your day. Gmail lets Claude search a client thread and write the reply as a draft; the connector drafts, I still hit send, and I want it that way. Google Calendar lets Claude check my week and create events when scheduling comes out of another task. Google Drive lets Claude search and read files, so a client doc becomes context instead of something I copy-paste in.
- 6
Connect the production layer: Make.com, Fireflies, Canva, Gamma, Semrush
Make.com is the big one for an automation business: Claude can list scenarios, read execution history when an automation breaks, and modify scenarios directly. Fireflies gives Claude my meeting transcripts, so a call can become a follow-up and task list without me re-listening to anything. Canva lets Claude search existing designs and export them, Gamma turns an outline into a presentation draft, and Semrush gives Claude live keyword and organic data when I am doing SEO work. Each of these replaces a tab I used to keep open.
- 7
Let memory build itself, then correct it
Claude Code keeps persistent memory across conversations: an index file plus topic files per project, stored in the .claude folder under your user directory. Mine currently tracks an ongoing client website audit, a live app, and an internal dashboard, none of which I wrote into memory by hand. Do not try to pre-write your memory files on day one. Work normally for a week, and when Claude states something stale or wrong, tell it to update memory in that moment. Corrections stick better than prefabricated context.
- 8
Run one real project end to end
Pick something real, not a tutorial app. The first project should exercise the whole chain: context pulled from your connected tools, code built in a repo, pushed with the GitHub CLI, deployed with Vercel. Mine include an AI career planning app and an internal ops dashboard, both built this way. The pass-fail test is simple: at the end, is something live at a URL that was not live before.
- 9
Prune what you do not use
Every connector adds tools to Claude's context, and unused ones are pure noise. After two weeks, look at which connectors you have actually invoked and disconnect the rest. The setup above is what survived that pruning for my business; your list will look different, and it should.
## hard-won tips
- !Connect tools you use daily, not tools you might use someday. The value comes from Claude reaching your real work, and an idle connector just clutters the toolset.
- !Keep the send button human. I let Claude draft client emails and Slack posts, but external communication gets my eyes before it goes out.
- !OAuth connectors mean no API keys in files for any of these tools. The only secrets on this machine are the two CLI logins, and those are handled by gh and vercel themselves.
- !A half-authed setup kills the habit. If every third request hits a login wall, you will stop asking. Finish the auth pass in one sitting.
- !Footer note for this page: godofclaude.com is an independent site. Claude is a trademark of Anthropic, PBC, and this site is not affiliated with or endorsed by Anthropic.
$ follow --the-build
Watch it happen, don't take my word for it
Every build on this site gets documented as it happens — the prompts, the dead ends, the results. No course at the end of this funnel. There is no funnel.
follow on x →