Connect your accounts to your agent

Connect this now. Your agent saves your work through GitHub from the next module onward, so this is the one connection everything else leans on.

GitHub is where your code lives. When your agent is signed in, it can create a repository, push changes, open pull requests, read comments, reply, and merge when you say so. That is the whole loop you will use later: draft, review, fix, ship.

Two routes show up today, and they work together. A command line tool is something the agent runs to do work. An MCP connector is something the agent talks to for extra read and write powers inside a service. Prefer the tool; add the connector when it exists.

Ask for enough access to do the job end to end. Tight scopes sound careful, but they trap you in repeat approvals. Safety here comes from three things: this is your personal account and your own private repos, the agent shows you what it will change before it changes it, and every change is versioned so you can undo mistakes.

Use a personal GitHub account for this course. If a work login wall appears in the approval tab, sign out there and sign back in with your personal account.

You can leave this until module 6, when you put your website online. Doing it now costs a few minutes and saves interrupting yourself later.

Vercel is where the marketing site will live. When your agent is signed in, it can create a project, link it to your GitHub repository, set environment variables, and trigger a deploy. Later, you will point a domain at it.

The flow is familiar: the agent installs the Vercel tool, starts login, your browser opens, you pick the account or team, and you approve. There is also an official MCP connector; take it so the agent can ask Vercel for structured status and logs.

If your Vercel account lives under a team, choose that team during approval so the agent can manage that project. You can move a project later, but it is tidier to choose the right place now.

Connect this now. Your database and your sign-ins both live here, and module 4 starts with your agent creating tables.

Supabase is where the app stores who signed up and what they saved. The tool here needs an access token you copy from the dashboard once, then the agent stores it safely. That unlocks creating a project, setting database settings, and managing auth providers.

There is an official MCP connector too. Add it so the agent can ask structured questions about projects or tables without scraping pages.

You will copy one token from your Supabase account page. That token is like a spare key: keep it in the agent’s safe place, not in an email or a note.

Only if you are building with React Native, which is what this course uses. Needed in module 3. If you take a different route to the app stores, skip this and connect that route's tools when you get there.

Expo builds your iOS and Android app in the cloud. The tool here is called EAS CLI, and the agent runs it. You sign in with your Expo account through a browser window that opens when the agent starts login.

There is no MCP connector for Expo today. The tool is what matters, because it can create builds, manage credentials, and upload to the stores later.

When the browser asks, pick the Expo account you created earlier. If you later add a team on Expo, the agent can switch to it when needed.

You can leave this until module 5, when your app starts sending email.

Resend handles transactional email like sign-in codes and receipts. There is no tool to install; you create one API key in the dashboard, and the agent saves it in the project’s secret file.

Resend offers an official MCP connector. Adding it lets the agent check sending status and templates through a structured channel.

You will create a key once, copy it, and paste it to the agent. That is it. Keep the key out of screenshots and notes; it is the same as handing someone a stamped stack of envelopes.

You can leave this until module 8, when you start taking money. Test keys only today either way.

Stripe is for payments. Two pieces connect here. First, your agent signs in with Stripe’s tool using a pairing code that you confirm in the browser. Second, you create TEST keys in the dashboard and the agent saves them in .env for the app itself.

This pays off later when you test checkout without spending money. Use only TEST keys now. Live keys come much later, when your app is ready and your business details are set.

Stripe also has an official MCP connector. Add it so the agent can check customers, prices, and events in a structured way.

You can leave this until module 7, when you move heavy work off the phone.

Modal runs server jobs for things your phone should not do, like image processing or long AI tasks. The agent uses a setup command that asks for a token; you can also create a token in the dashboard and paste it once.

There is no MCP connector for Modal. The tool is what matters here, because it can deploy and run jobs on demand.

You will see a one-time token screen. Treat it like the spare key under the mat: pick it up once, hand it to the agent, and then do not leave it lying around.

Optional, and not urgent. It reviews your code once other people are touching it. Plenty of people ship without it.

CodeRabbit reviews your pull requests and leaves comments. This one is different: you do not give a key to your agent. You install CodeRabbit on your repository from its own site as a GitHub App.

The flow is all in your browser: click Install, pick the repository, and approve. Once installed, any new pull request will get a review from CodeRabbit without the agent doing anything special.

If you do not see it on a first pull request, open the app’s settings and confirm the repository is selected.

A secret is like a spare key. You would not glue it to the front door. Put it in one place only: a file named .env at the root of your project. The agent writes and reads this for you. You never paste secrets into code files.

The app code reads values like RESEND_API_KEY or STRIPE_SECRET_KEY from .env when running locally, and from the host (Vercel, Modal, Expo) when deployed. The agent will set those host copies later so your builds still have what they need without baking secrets into the app.

If you ever need to move a project, the agent can show you what is in .env and help you re-enter those values into the new place. Keep a secure copy of the values in your password manager, not a sticky note.

Git ignores files listed in a file named .gitignore. That is the do-not-pack list. .env must be on it, so the agent can use your secrets locally while GitHub never sees them.

Ask the agent to prove it. It will show you the line that keeps .env out and confirm that your last commits did not include it. If anything is wrong, the agent will fix the ignore list and clean any accidental history safely.

Leaks happen. Treat it like losing a house key. You do not replace the lock; you change the key. Every service gives you a “revoke” or “delete” button for a key and a way to create a fresh one.

You will click revoke in the dashboard, generate a new key, and paste it to the agent so it can update .env and any host settings. That is normal enough that dashboards make it a two-minute job.

Nothing to do in this section. You will see these three letters in your agent's replies and in other people's advice, and it is easier to meet them here than to wonder later.

A command line tool, or CLI, is a small program your agent runs on your machine, the same one a developer would use. It types a command and reads what comes back. Almost every service has one, and it is what did most of the work in the sections above.

An MCP connector is a direct line between your agent and the service, so it can ask a question and get a clean answer instead of running a program and reading the output. MCP stands for Model Context Protocol. Only some services have built one so far.

Your agent set up whichever were available while you were approving things, and it picks between them on its own. You will not manage either by hand.

Before moving on, get one clear report. The agent should confirm which accounts it is signed in to and which keys it has stored, then point out any missing pieces and tell you how to finish them.

This is the moment to catch small gaps: the wrong Vercel team, a missing Supabase token, or Stripe keys not saved. Better to fix them now than while deploying later.