Multi Agent Orchestration with MCP and A2A

An early-stage project exploring multi-agent orchestration with the Model Context Protocol (MCP) and agent-to-agent style coordination.

The current focus is a local MCP server that connects to Claude Desktop and can run terminal commands in a controlled workspace.

Project Status

Status: Early development
Phase: Foundation setup and MCP integration
Current focus: Claude Desktop connection, terminal tool behavior, and project tracking

Related MCP server: Claude Desktop Commander MCP

What This Project Is About

This repository is a practical experiment in combining:

• Model Context Protocol for tool integration

• Multi-agent orchestration ideas

• Local command execution through a custom MCP server

• Claude Desktop as the front-end interface for interacting with tools

The goal is to build a small but useful orchestration setup that can grow over time.

Current Highlights

• MCP server created using FastMCP

• Claude Desktop config connected to the local server

• Terminal command execution tool being tested

• Working directory issues identified and fixed

• Project progress and fixes tracked in dedicated log files

Project Structure

• mcp/servers/terminal_server/terminal_server.py - MCP server for running terminal commands

• FIXES.md - Detailed log of fixes, file changes, and reasoning

• PROJECT_STATUS.md - Progress tracker and GitHub push history

• main.py - Project entry point or placeholder script

• pyproject.toml - Python project configuration

• uv.lock - Locked dependency state for uv

MCP Server Overview

The current MCP server exposes a terminal command tool that:

• accepts a command string

• runs it through subprocess

• returns command output or errors

• uses a stable default workspace on the Desktop

This was added so Claude Desktop can interact with local commands without relying on a fragile working directory.

Setup

1. Create the virtual environment

uv venv