I Made Claude Code Talk to My Grok Bot (and Control It)

0:00I just had Claude code ask my Grok bot to search X. A minute and a half later, the top five posts, links, views, the lot. It didn't touch the X API, and it cost me nothing extra. [music] Grok bot won't let you pick your model. Elon Musk said Grok bot will use the best model for each job. Claude Opus 5.5 is one of them. Sounds good, but you

0:24don't get to choose. Cursor's own help page says it plainly, there's no model picker. And your usage just shows up as Grok bot usage. So you can't even tell when you got Opus. A lot of us already pay for Claude or ChatGPT. How do you get those models to work with Grok bot? Later on, I'll get Grok bot to write with Opus just by

0:43asking for it. There are two ways to do it. Way one, Grok bot is the boss. It runs Claude code or Codex on its own computer. Way two, Claude code is the boss. It sends jobs to your Grok bots and gets the answers back. A few people have covered way one already. Way two, not many people know

1:04about. So I'll start there. [music] Cursor has something called the agent SDK. It comes with an official add-on that lets your agent message the Grok bots on your account. It's the same bot with its memory and its tools. You send it a message, it does the work, and the reply comes back into Claude code. And it's not just Claude code, Codex can use it, too. And

1:27honestly, so can any other LLM. What you need: Grok bot on your plan. On the new X subscription, that's premium, $30 a month, or one of the older Cursor Pro plans. It signs in with your Cursor login and uses your Grok bot weekly allowance. No API key to buy. One heads-up. On the 23rd of October, Cursor removed this add-on from the newest version of the SDK. It still works on the version just before. My free guide is set to that version. So, if you want to keep using this, don't update it, and it could stop working

2:01anytime. [music] Setup is three steps. One, download my relay from the free guide and type in your bot names exactly the way they show in the app. Two, log in with your cursor account. Three, give Claude code a small scale that sends the message and reads the reply. Mine is in the free guide below.

2:23You can copy it. Then, you just ask Claude code in plain English. So, here I'm asking Claude code for the five most viewed posts this week about Kling four, the new AI video model. Look at what lands in Grok Bot. I typed one line, Claude wrote X, a proper brief. Research only, don't post anything. This week's dates. What to send back for each post. And if you can't confirm a view count, say so,

2:48don't guess. First, X says it's searching. The actual results come as a second message. Claude code waits for that one and picks it up. It brings back X's list word for word. Then, it adds its own take. The top post from Higgs Field has 3.6 million views, about 50 times the next one. Claude flags that it's probably paid promotion, so don't read it as real interest. X also spotted two posts with almost the same launch copy, likely a planted

3:17creator push. [music] So, Grok Bot does the X searching, Claude does the thinking. I use this every week for my videos and for general research for my business. [music] The X API charges for every post it sends back, half a cent per post. One search that brings back 100 posts, 50 cents. Do 50 searches a day for a month, that's

3:39$750. Through Grok Bot, it's part of your plan, nothing extra. Some people do this with a webhook and a secret key instead. That works, too. But, if the key leaks, someone else can run your bots. This way just uses your own login. [music] It's not perfect. These are the problems I hit. One, spell your bot name wrong and it quietly makes a brand new empty bot. Check the

4:06name before you send more work. [music] Two, if the bot is still busy with an earlier job, your new message just doesn't go through. Wait for it to finish, then ask again. [music] Three, the reply comes in two parts, searching now then the actual results. Make sure you read the second one. [music] Four, on Windows, the command line strips out the quote marks and the message breaks. Run it through Node.js

4:30instead. My skill already does that. Five, your login expires once in a while. Just login again. Now, the other way around. Every Grok bot has its own computer and that computer has a terminal. So, you can install Claude Code and Codex on it and sign in with your own plans. You only sign in once. For Codex, turn on device code sign in first. It's in ChatGPT under settings, security and login. If you skip it, the login can

5:00fail without telling you why. Here's a quick social post start to finish. First, the picture. I asked my image bot to use Codex. So, instead of Grok's own image model, it uses ChatGPT's on my ChatGPT plan. About a minute later, there's the puppy. Then the words. I asked for Opus 5.5 by name to write Facebook and Instagram captions. It runs through Claude Code on

5:25that computer. Two options for each with hashtags, wet fur, dry tail, zero regrets. [music] Remember, Grok bot won't let you pick Opus. Here, I just asked for it and got it. Is this allowed? I checked. Anthropic's own page says you can sign in to the official Claude Code with your own subscription, even on a hosted computer

5:46like this. It gets gray when a bot keeps your login to share with other people. Don't do that. One more safety thing. Point Claude code at its own project folder. In auto mode, it can edit files and run things on its own. You don't want it touching everything on that computer. [music] So, which one should you use? For quick everyday questions, just use

6:06Grok bot on its own. If you want a specific model and you're on your phone, use way one. Grok bot runs Claude code or Codex for you. If you already live in Claude code or Codex, use way two. Grok bot becomes your helper with X search and all your bots for nothing extra. I put the setup steps, my skill, and the five fixes in a free guide. The link is

6:30below. No email needed. [music]