Setting up Grok Bot takes about ten minutes, but the order you do things in changes how useful it is by day three. This walks through the actual setup: your first bot, the plugins worth connecting first, and the specific mistakes real early users ran into before finding a working routine.
A note on scope: this covers getting a working team running, not whether you should connect it to anything sensitive. Shared sessions across bots, what happens with banking or payroll credentials, and what SpaceXAI's own documentation says about liability are covered in full in our companion article on what Grok Bot actually is. Read that first if you have not already, since the credential sharing described below is real and by design, not a bug. For pricing changes, new availability, or anything that has shifted since launch, see our Grok Bot news and updates tracker, refreshed weekly.
Grok Bot does not hand you a blank chat box. Every plugin you add, every bot you name, sticks around and works while you are doing something else. That is the point, and it is also why the order matters more than it looks like it should.
Before you open the app
Access runs through a Cursor account, not a standalone Grok subscription: pricing sits on the Cursor Ultra or Cursor Premium Teams plans, detailed in the companion article above. Rundown Guides' own setup tutorial points new users toward Cursor's seven-day free trial as the entry path, and AI educator Debbie O'Brien, who documented her full setup on video, reported using most of that trial window in two days of active testing. Treat the trial as a real constraint, not a formality.
Have ready before you start:
- A Cursor account
- The apps you actually want a bot working in: Gmail, GitHub, LinkedIn, Notion, Slack, and Google Calendar cover most first setups
- Ten to fifteen minutes uninterrupted, since the first bot walks you through a short onboarding conversation
Start with one bot, not a team
Every early setup that worked well started the same way: one bot, the premade Chief of Staff role, before anything else gets built.
The reasoning shows up independently in three separate places. Krista Letz, SpaceXAI's own enterprise GTM lead, published a playbook centered on a Chief of Staff that orchestrates specialists rather than working alone. Debbie O'Brien, in the video walkthrough of her own setup, put it plainly: name one bot, let it look at your connections and your work, then ask it to design the rest of the team for you. Rundown Guides' tutorial, built around the same pattern, calls it a focused team led by one coordinating bot, and warns against spreading a new user across many bots on day one.
The prompt O'Brien used to get her clearest results asked the Chief of Staff to review her connections, build a profile of her workstreams and projects, and then recommend the three most useful agents, automations, and plugins to add.
Answer its follow-up questions honestly. It uses them to decide which specialists are worth creating, not just to make conversation.
Connect your plugins, in the right order
Plugins live under Settings, then Plugins, in the desktop app. Add, authenticate in your browser, done, and every bot on your account can use it from then on: a connector installed once is available to every bot you create afterward, not just the one that installed it.
Two connectors behaved differently from the rest in O'Brien's setup, and it is worth knowing before you hit them:
GitHub can show as already connected and still block you. O'Brien found that with her GitHub linked through Cursor, a bot treated it as available for cloud-agent tasks immediately. But a second, separate GitHub connector inside Grok Bot's own plugin list still asked for a personal access token on top of that. If you hit this, it is not a broken connection, it is two different integration paths that do not share state yet.
X does not run on the same one-click flow as the rest. Where Gmail, Calendar, and Notion connect through a standard authorization screen, O'Brien found that X asks for a developer bearer token before a bot can post or read on your behalf. Budget ten minutes for the X developer portal the first time, or skip that connector until later and add it once the rest of your setup is working.
For anything without a plugin, the bot hands you its cloud computer directly: a banner appears, you sign in or clear a two-factor prompt yourself, then tell it you are done and it continues. It never sees the password itself in that flow.
Let the Chief of Staff build the rest
Once your first bot has looked at your connections, the fastest path to a working team is not naming five bots yourself. It is asking the Chief of Staff to propose them, reviewing the list, and approving what is actually useful.
In O'Brien's setup, a typical recommendation set for someone juggling code, content, and outreach looked like: a coding bot scoped to specific repositories, an inbox bot for triage and drafts, and a bot for one social platform. Add a fourth only when a real task is genuinely getting in the way of the first three, not before. O'Brien, after building out seven bots of her own, put it simply: do not add a bot for every subtask, or you are just adding another handoff to manage.
Rundown Guides notes that each bot gets one conversation thread by design. That is a structural reason to keep bots scoped to a role rather than piling every task into a single thread and hoping it stays organized.
Five things that will trip you up, all in one place
None of these are documented together anywhere else, which is exactly why they are worth collecting here instead of leaving you to find each one the hard way.
- Posting to LinkedIn is one-way once it is live. O'Brien found that a bot can draft, review, and publish a post, but LinkedIn will not let an already-published post be edited to add an image. If you want a screenshot or graphic attached, ask the bot to add it as the first comment instead, not as an edit to the post itself.
- Frequent connector syncing costs more than it saves. Rundown Guides warns that asking bots to check and re-sync a connected service constantly burns through usage faster than batching updates. Save-and-report once at the end of a work session, rather than continuously, is the pattern that held up.
- A GitHub connection through Cursor does not automatically satisfy Grok Bot's own connector. See above: do not assume one green checkmark covers both paths.
- Publishing takes longer than a normal social post. In O'Brien's setup, several minutes passed between asking the bot to post and the post actually appearing. That is expected behavior, not a stall, but it is worth knowing before you assume something failed and try again.
- A seven-day trial goes fast under real use. O'Brien reported using most of a week-long trial in two days of active testing across multiple bots and apps. Plan your first real session accordingly rather than spreading it thin over a week.
Three prompts to copy on your first day
These come from Bot Directory, an open, community-maintained prompt directory for Grok Bot where anyone can contribute a tested setup:
Chief of Staff briefing. Connect the tools you use daily, then ask it to walk you through connecting each one and to save itself as a bot once it has what it needs.
Account-specific tracking. For anyone in sales or customer-facing work: a bot dedicated to one account, watching the relevant Slack channels, digesting calls and emails as they happen, and answering questions about that account from the context it has built up over time.
Recurring account digest. A weekly-running version of the above, useful once you are managing more than two or three accounts and want a standing summary instead of a one-off setup.
Start from one of these rather than a blank prompt. Adjusting a working setup is faster than describing what you want from scratch, and every prompt in the directory has already been tested by someone else first.
What this means if you're getting into AI engineering
Setting up a working agent team surfaces the same questions that show up at production scale: how you scope permissions, how you decide what runs automatically versus what waits for a human, and how you structure a system so one part failing does not take down the rest. That is exactly the layer that 4Geeks Academy's AI Engineering program trains for.