MCP Agent Coordinator

You can run and coordinate multiple AI agents against a single codebase with the Agent Coordinator MCP server. It prevents conflicts, manages dependencies, and provides clear, organized parallel development so your agents can work together efficiently without stepping on each other’s changes.

How to use

You connect your MCP client to the Agent Coordinator MCP server to assign work to AI agents. The server exposes a single HTTP endpoint that your client subscribes to for updates and to receive new tasks. Agents request work, claim a todo item, and the server locks the relevant files to prevent conflicts. When an agent finishes, the server unlocks the files and triggers the next dependent work item.

Use the system to organize work into Projects, Tasks, and Todo Items. Each Todo Item represents a non-conflicting unit of work. The server automatically routes available work to idle agents, ensures dependencies are respected, and provides real-time visibility into progress.

How to install

Prerequisites: you should have Python installed on your system. A Python virtual environment is used to keep dependencies self-contained.

Step 1: Prepare your environment

• Ensure Python 3.x is installed and available in your PATH.

Step 2: Set up and run the server

• On Windows, run the provided startup script to create the virtual environment, install dependencies, and start the HTTP server on localhost:8001.

Step 3: Verify the server is running

• The server exposes its status at http://127.0.0.1:8001/sse and will be ready to accept MCP client connections.

Configuration and usage notes

Configure your Cursor MCP settings to point to the Agent Coordinator MCP server. The connection should use the SSE endpoint where updates and tasks are streamed.

Cursor configuration example (paste into your Cursor MCP settings):

{
  "mcpServers": {
    "agent-coordinator": {
      "url": "http://127.0.0.1:8001/sse"
    }
  }
}

Security and workflow notes

The Agent Coordinator manages work with centralized ownership and non-conflicting assignments. It tracks progress, prevents duplicate work, and enforces dependency order so you can scale agent collaboration without chaos.

Troubleshooting and tips

If your MCP client cannot connect, verify that the server is running and listening on localhost:8001. Check that the SSE endpoint is accessible at http://127.0.0.1:8001/sse, and ensure any firewalls allow traffic to that port.

Notes on usage goals

This system is designed for true parallel development where agents can work concurrently on separate Todo Items without conflicting edits. It utilizes file locking, task dependencies, and centralized progress tracking to keep the project coherent.