Empowering LLMs with MCP Tools

Building MCP tools is very niche and modern technology. It’s been handful of days since the Claude host (Cline) been released. Microsoft, is catching up too with VsCode extensions (CoPilot for GitHub). In addition, custom ones have been built by the community (open sourced) such as MCPHub and MCPCli.
All these tools use the same architecture to good extend. The below architecture demonstrates how users interacts with the client and receive accurate output. Traditionally, LLMs would hallucinate as text generators due to inability to execute commands or access real-time data. MCP empowers LLMs to execute code.
MCPs are like the robot arms with the LLM being the brain.
I’ve created Parrot
, which is MCP for repeating the user’s input. while this MCP doesn’t do much, it’s built as a proof-of-concept.
In addition, I’ve created two other MCP servers in Pything + fastAPI, and C#.
These servers are configured centrally in JSON. This is used as an input by the host (Cline for eg.) to run each tools using the provided command.
It is possible to set the settings to dictate a trigger command. Moreover, It can be configured to wait for the user’s permission before it runs. This can be useful for mission critical applications.
NeoVim is also empowered with native Lua MCP tools for local development for accessing the file system. For example, executing ls
for listing the directories files (see image below).
Cline use them for managing the project’s files. Also, vibe coders benefit from these tools directly, without knowing!
Here’s the sample input. (Credit to echo on the MCP-Hub)
and sample output
The C# Tool’s output.
These tools can run universally on any standard MCP Hub.
Have a question, don’t hesitate to contact.