If you’ve been into vibe-coding, or even been using AI for generating code and helping you to build software, you might have thought to your self, “it’s really annoying just sitting there waiting for the agent to finish”. It can sometimes take up to a couple of minutes just waiting for the agent to finish, and what do you do when you’re just sitting there? The whole point of using AI to help us build stuff is to save time, so just sitting there and waiting doesn’t make much sense. Discobot solves this by letting you run multiple AI coding agents at once — and if you’re building on top of MCP, it pairs well with the broader ecosystem of MCP tools for developers.
With Discobot, you can run multiple AI coding agents at once across isolated sandboxes. When you have instructed your agent to execute a task, just open up a new session, and make the new agent do something else.
In this article, I will explore what Discobot can do for you, and also use it in an existing GitHub repository for a hands on demo.
Once you’ve installed it, you need to configure a coding agent like Claude Code. For more on managing MCP servers in Obot, check out our step-by-step guide.
If you’re not automatically presented with a modal where you can configure Claude Code, then click this “cog wheel” icon up in the right corner:
This should open up a new modal where you can sign in to Claude/Anthropic:
When everything is ready, you will be presented with a screen like this:
Feel free to play around as you like. You can try out a sample project, or you can use your own. Come back here when you’re ready to continue.
Try Obot Today
⬇️ Download the Obot open-source gateway on GitHub and begin integrating your systems with a secure, extensible MCP foundation.
Testing out Discobot
The repository I’m going to be using will be a simple Django blog. I’m going to try to create a new contact page, and also implement a comment system for the posts.
I will first start by asking Discobot to implement a comment system, this can take a little while. And when this is running in the background, I will start a new session where I ask Discobot to create a simple contact page.
I have set up the project locally, so I click “Use Existing Project on Disk”. Inside the modal that opened up, I type in the path to my local project and click “Add Workspace”.
When that is done, I am presented with this screen, where I can just start typing in the text area, and hit enter to instruct my agent to start working on my project.
For this demo, I wanted the agent to build a commenting system for the posts. The prompt I used was simple and straightforward — just a plain-English description of what I wanted.
Usually, I would now just sit and wait for the agent to finish its work. But since I can just open up a new session, that’s exactly what I did.
I clicked “+ New Session” on the top of the screen, and a new session started. In this session, I wanted the agent to start working on my contact page.
If I wanted to now, I could start even more sessions and have more agents running. But two agents should be enough for this demo. I could also run a different workspace as well if I wanted to.
Now I go back to the agent for the comment system to see the status.
When you create new models in Django, you need to update the database, and for this to work, you need a python environment. I have actually never seen an agent do something like this before, but it created a new environment and knew what to install for this to work.
After a little while, the environment was set up, and the migrations were running as intended.
At this point, everything in the comment system was ready.
Up in the left corner, you’ll find a button saying “Terminal”. Click this to get a terminal for the project. Here, I was able to see the coding changes that the agent had done for me.
I was now able to see all the changes and updates, and I verified for my self that everything I wanted it to do was done. If there were any small changes I wanted it to do, I could just go back to the “Chat”, and then ask the agent to do so.
Previews
Let’s say I wanted to see in a browser what has actually happened. I can instruct Discobot to create a service for running my project, and I will get a URL to see it in a browser.
I first just need to instruct the agent to read some documentation. So in the chat, I do this:
Nice, now it knows a little bit more about how to do the next step which is to create a service:
When I did this, a new button appeared up in the left corner:
You can see there is a new button “Django Dev Server”. If you click this, you’ll be taken to the service we just created in the previous prompt. Here you can see the comment form:
Next step then was to check out the changes that was made for the contact page.
Since individual sessions are isolated from each other, the changes for the comment system was not implemented here. Only the changes for the contact page, and this makes it much easier to see what has actually been done.
A Third Session
Let’s say that these two sessions are done, but there is one more thing you want to do in your project. Maybe one of your users has reported about a bug or similar.
This will then start yet another session and start working on a solution to help you with this. After a little while, I got the result back with an explanation on what it had done:
This sounds like a good solution when I read the code, but I still wanted to see a new preview. So I started up the service again, and here’s the result:
And this looks much better than what it did. It still needs some work with the spacing, fonts, and colors, but that’s out of the scope for this article. If I wanted to though, I could just go back to the chat and ask the agent to fix this for me.
Editing the Code in an Editor
Let’s say that I like everything the AI has done, but there are some small changes I want to do, and I want to just do it manually. This is possible through SSH, and there is a very easy way to accomplish this.
Up in the right-hand corner, you will see something like this:
I love using Cursor, so in this list, I just clicked “Cursor”, and my editor opened up. You’ll probably get a question if you trust the folder, and I clicked yes on this.
After a few seconds, I had my project open in Cursor, and I could do changes manually:
Summary
As you can see, Discobot is a powerful tool for anyone who wants to run multiple AI coding agents at once. In this walkthrough we covered:
Running parallel agent sessions across isolated sandboxes
Live browser previews of agent-built features
SSH access and manual editing via Cursor
Starting additional sessions for bug fixes without interrupting ongoing work
If you’re ready to go further, check out 15 of the most useful MCP servers to extend what your agents can do. Stay tuned for more things you can build with Discobot.
