Agent-MCP is a framework for orchestrating multi-agent AI systems, designed for experienced developers building complex AI applications. It addresses limitations of single-agent approaches like context overflow and lost knowledge by enabling specialized AI agents to collaborate efficiently through a shared context and intelligent task management.

How It Works

Agent-MCP breaks down complex development tasks into linear, manageable steps, allowing specialized agents to work in parallel. It utilizes a persistent knowledge graph (RAG) to store all project context, ensuring agents have access to relevant information without context pollution. This approach, termed "Ephemeral Agents with Shared Memory," contrasts with traditional long-lived agents by using short-lived, focused agents for specific tasks, enhancing security, performance, and predictability.

Quick Start & Requirements

• Install: Clone the repository, set up environment variables (.env with OpenAI API key), and install dependencies using uv venv && uv pip install -e ..
• Prerequisites: Python 3.10+, Node.js 18.0.0+, npm 9.0.0+, OpenAI API key. Claude Code or Cursor AI assistant is recommended.
• Run: Start the MCP server with uv run -m agent_mcp.cli --project-dir /path/to/your/project. Launch the dashboard from agent_mcp/dashboard with npm install && npm run dev.
• Docs: Getting Started Guide, MCD Creation Guide, Theoretical Foundation.

Highlighted Details

• Real-time visualization of agent collaboration via a dashboard.
• File-level locking to prevent agents from overwriting each other's work.
• Specialized agent modes (e.g., --worker, --playwright, --memory) for tailored behavior.
• Strict lifecycle management with agent termination and cleanup protocols.
• Focus on linear task decomposition for atomic, testable steps.

Maintenance & Community

• Active Discord community for support and discussions.
• GitHub Issues for bug reports and feature requests.
• Open to contributions with a clear contributing guide.

Licensing & Compatibility

• MIT License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

The framework is designed for experienced AI developers and requires familiarity with AI coding workflows and distributed systems. Documentation is actively being improved.