So there is a lot of buzz about MCP (Model Context Protocol) servers and rightly so. They're really interesting and powerful tools. The most interesting I've found so far was the use of them to break up large tasks for AI agents.
Maybe this is not a shock to anyone, but I use copilot a lot on this site. Not to generate text content. Honestly, I don't think it really sounds like me and I find that a bit odd. But it can be really handy to multi-task for different projects. I'm also a new user of Warp, which is a pretty slick terminal with agentic AI built in. There is a lot of ground to cover in the world of AI tooling, but we're going to just focus on MCP servers today.
MCP servers allow you to extend the functionality of your AI agents. I think its best to think of it like a toolbox for them to use for specific tasks or resources. It's basically a layer that sits between tools and resources, and the agent itself. A common protocol to allow different tools to communicate with the agent.
So on a whim, I decided to make something that is basically a copy of Task Master AI, and by that I mean, a vastly inferior local file version 🤣. But the goal was to really get the basic construct down and how then to integrate it into VS Code. It was also because I wanted a proof of concept for how I might make more that server more specific purposes for me (so for real, they have a great tool and I'd recommend it).
Because I'm a Tolkien nerd, my tool is called...The One Ring. Well technically its called "one-ring" because I think Sauron would've understood that articles and spaces just aren't great for file system paths. But the name does have a certain ring 💍 to it, doesn't it? Ok...I'll stop 😉.
It does kind of fit though, I'm trying to bend this agent to my will by breaking up larger complicated tasks so that the context doesn't get all fragmented. Of course, I used Warp to build this, in particular I'm a big fan of Warp's FFWD mode (I'm not sure that's what its called but that's what it seems to do). You can just let it run and it'll keep churning on an idea. So I gave it a PRD (Product Requirements Document) and let it go.
The important thing to remember is that basically you're building a server to expose resources, tools, and prompts. Prompts are how to give the agent a way to communicate actions or interact with your server. Resources are just that, data, or information loading. Tools are the ways to execute actions or create side-effets. Common types of MCP server are for things like web crawling, task management (like this one), code assistance, infrastructure helpers, testing tools, and more.
So my project is just a standard IO MCP server. I just wanted to spin something up locally for the moment and see how it went, but the instructions are all in the repo. Feel free to check it out. For vscode, you can basically use the vscode-settings-template.json in the root of the project. Of course, you won't want to use my file paths 😆.
1{
2 "mcp.servers": {
3 "one-ring-task-manager": {
4 "command": "/Users/<user>/<code_dir>/code/one-ring/start-server.sh",
5 "args": [],
6 "cwd": "/Users/<user>/<code_dir>/code/one-ring",
7 "env": {},
8 "transport": "stdio",
9 "description": "One Ring Task Manager - AI project and task management MCP server"
10 }
11 }
12}Feel free to use this repo as a template but honestly the ability to ask agentic AI to help you build them is pretty powerful. This didn't take me long to generate, even as a proof of concept. I may continue to refine it as it gets more use, but heck Mordor wasn't build in a day.
