New Monti APM MCP

I saw a short demo for a Sentry MCP Server a couple of days ago and thought “Hey, I wan’t that for Monti APM”. So today I made Warp build one.
So it’s really fresh and not tested with anything besides Warp itself. It’s already used that new Tool to help with debugging stuff we were working on later this day.

7 Likes

Really cool, @janmp!

Was it easy to build it? It took only a day?

It’s completely vibe coded with warp.dev and took about 2 hours. Vibe coding MCP servers works fine. They are usually relatively small and just have to do a few tasks and you need an agent to test them anyways.

And it’s quite fascinating to have AI write its own tools. When I first tried doing that I used Claude Desktop (with the desktop commander MCP, to allow it to read/write files etc). Claude Desktop with its standard setup has a really bubbly and excitable personality. I was experimenting with having Claude use my obsidian vault at the time and suggested we could maybe access the smart connections obsidian plugin to give Claude semantic search of my vault. It did some research and found a way and then got really excited, wrote a plan how to do it and set it as a top priority task on my kanban.

1 Like

@zodern should know about this.

Yeah, I had planned to include him in the announcement but then forgot. There’s probably quite a bit room for improvements.

Nice work Jan-Claude :clap: . So what sort of things have you been able to do it? Can you share any example usage ?

Quite frankly, I have not really used it since I have been testing it. I am still in the process of figuring out how to use warp.dev efficiently. The workflow is completely different from coding by hand. You do less work coding yourself but you pay for that by having to do context engineering, so your agents know how you want them to do things. I also want to take care of side effects of this type of work. You don’t really know your code as well when you have your agent write it and I think you also loose your edge overall much faster than you think.

Overall it seems like with a setup like warp.md the coding agents are not really eager to use MCPs at all. Warp started out as a terminal app with just a little bit of AI in the beginning (it had some terminal input completion, wich is already very helpful) and developed into a coding a full fledged coding agent solution. But except for a few specialised tools to e.g. handle code divs it does all that with terminal commands.

And there are some real advantages to that. E.g. I added BrightData tools to extract web content. But with bright Data the LLM gets the complete content for the url jammed into its context. BrightData is good for use in Agents that need to extract websites that get blocked by popups and captchas, but you don’t have that as much in resources for coding.

With using plain old curl warp can do something like this (to find out the current Meteor Version in this case):

curl -s https://install.meteor.com/ | grep -o 'RELEASE=.*' | head -1

and it get’s exactly the answer is looking for.