Agent-MCP provides a robust multi-agent collaboration framework built on the Model Context Protocol (MCP). It exposes a suite of tools to create, manage, and coordinate specialized AI agents that work together in a persistent knowledge graph, visualize agent activity, and orchestrate tasks across codebases. Once connected as an MCP server, clients can dispatch actions such as creating agents, assigning tasks, querying project context, and managing inter-agent communications. The included MCP tools enable orchestration of parallel work streams, monitoring of agent status, and querying or updating the shared memory graph to preserve architectural decisions and implementation details across sessions.

Prerequisites:

• Python 3.10+ and Node.js 18+ (per the Quick Start recommendations)
• uv (Uvicorn/uv) or your preferred Python environment manager installed
• Git

Installation steps:

1. Clone the repository: git clone https://github.com/rinadelph/Agent-MCP.git cd Agent-MCP

2. Set up Python virtual environment and install dependencies (recommended): uv venv uv install

3. Obtain an OpenAI API key and configure environment: cp .env.example .env

   Add your OPENAI_API_KEY in .env or export it in your shell

4. Run the MCP server (example using uv): uv run -m agent_mcp.cli --port 8080

5. Optional: Launch the accompanying dashboard if you want a GUI for monitoring agents: cd agent_mcp/dashboard npm install npm run dev

Notes:

• The README also provides a Node.js/TypeScript alternative setup inside agent-mcp-node if you prefer a Node-based stack.
• If you’re using a Node-based workflow, you can adapt the commands accordingly (see Node.js implementation in the README).

Tips and considerations:

• Ensure OPENAI_API_KEY is available in the environment where the MCP server runs.
• The MCP server in this project is designed to expose endpoints over HTTP and WebSocket as described in the MCP integration guide (http://localhost:8000/mcp and ws://localhost:8000/mcp/ws by default).
• For Claude Desktop or other clients, configure your MCP server’s transport (HTTP or WebSocket) and verify the host/port in the client config.
• Environment variables like AGENT_MCP_HOST and AGENT_MCP_PORT can be customized to fit your deployment scenario (e.g., containerized or cloud environments).
• If you modify the server port, ensure clients are updated accordingly to avoid connection failures.
• When using dashboards, ensure npm dependencies are installed in the dashboard directory (agent_mcp/dashboard) to enable the visual UI.