MCP Multi-Agent Modular Awesomeness!

Overview

This is a primative multi-agent modular initialization capability test prototype of Model Context Protocol. My test project will have multiple servers and clients. They will operate in a modular way, the first will create a markdown file based on adding user input text to a content template and saving the file to root directory. The second will ingest that file and add more info to it then save the updated version to the root. The alt A third will likewise ingest the output of the second and add still more content and export the third version of the file to the root. The alt B third server alternatively could take another action. Treat each of these three like back boxes in the sense that they will not share code or components but we will pretend they are running on different technologies to preserve the modular approach and we will for convenience be able to do things like share a virtual environment and requirements for this first test. The second server will be a decider and will choose which is the best next server.

The notion of strict modularity such that different MCP interoperable system could be created and operated by totally different teams or even different companies is potentially a key unlock for MCP and for LLM based agents to take off and be scalable and more fully userful. This is described more fully in the project overview for “Agent” here: https://github.com/dazzaji/mcp-project/issues/1.

Let’s Run the Project!

(Based on continuing periodic issues, I’ve further documented below how best to run and use this project based on the various obstacles that have popped up. Note that the file paths should be updated to your machine and I hard-coded the version of Python I’m using to deal with countless issues pointing to the right version and maintaining the link to it.)

Okay, let’s get the mcp-agent-router project running with Claude Desktop and the MCP Inspector. Here’s a step-by-step guide that addresses your specific concerns, using uv as requested and focusing on understanding how to integrate Claude Desktop and use the Inspector for troubleshooting.

1. Project Setup with uv

First, ensure your project environment is correctly set up with uv.

# Clone the repository if you haven't already
git clone 
cd mcp-agent-router

# Remove existing virtual environments if any
rm -rf .venv venv

# Create a new virtual environment with Python 3.11
/Library/Frameworks/Python.framework/Versions/3.11/bin/python3.11 -m venv ./venv

# Link venv to .venv for uv
ln -s venv .venv

# Activate the virtual environment
source ./venv/bin/activate

# Confirm Python version
python --version
# Expected output is: Python 3.11.x

# Install necessary dependencies using uv