agent-orchestration

Agent Orchestration is an MCP server that enables multiple AI agents to share memory, coordinate tasks, and collaborate across IDEs and CLI tools. It provides a shared memory store for context, a turn-based task queue, resource locking, and real-time coordination status, making it easier for a team of agents and tools to work together on a single project. The system also auto-syncs important context to activeContext.md for easy reference, and includes an agent discovery layer so you can see who else is working on the project. The available tools cover session management, agent management, shared memory, task management, and coordination, enabling a complete orchestration workflow from initialization to task completion. This server is designed to work with a wide range of AI coding agents that support MCP or AGENTS.md formats, including popular tools like OpenAI Codex, VS Code Copilot, GitHub Copilot, Cursor, Jules, Windsurf, and more.

To use it in your project, you typically run the server via npx, pointing at the agent-orchestration package. The Quick Start shows how to initialize the AGENTS.md file for any agent, and how to set up Cursor-specific rules when using the Cursor IDE. The command surface includes init, init-cursor, serve, and help, enabling you to bootstrap your orchestration, start the MCP server, and inspect available commands.

Prerequisites:

• Node.js 18+ (and npm/yarn)
• Internet access to fetch the MCP server package via npx

Step-by-step installation:

1. Install Node.js 18+ from https://nodejs.org if you don’t already have it.
2. Verify installations:

node --version
npm --version

3. No global installation is required thanks to npx. You can start the MCP server directly using npx:

npx agent-orchestration serve

4. (Optional) If you want to preinstall for offline usage or faster startup, install the package locally in your project:

npm install --save agent-orchestration

Then run the server with:

npx ./node_modules/.bin/agent-orchestration serve

5. If you need Cursor-specific setup, initialize for Cursor which copies Cursor rules:

npx agent-orchestration init-cursor

6. Create and customize your MCP configuration in ~/.cursor/mcp.json (or your preferred MCP config location) to point to the agent-orchestration server as shown in the examples.

Tips and common considerations:

• The MCP server maintains per-project state (memory, tasks, locks) often stored in a per-project workspace. Ensure your project directory is accessible and writable by the process running the MCP server.
• Environment variable MCP_ORCH_SYNC_CONTEXT controls whether activeContext.md is auto-synced. Enable it if you want automatic context updates; disable if you prefer manual control.
• Use the provided tools to manage memory (memory_set/memory_get), tasks (task_create, task_claim, task_complete), and coordination (lock_acquire, lock_release, coordination_status) to avoid race conditions and duplicate work.
• The Quick Start instructions assume you’re running in a project folder and using npx. If your environment blocks npm/npx network access, consider local installation or a container-based approach.
• If you run into server startup issues, verify Node.js version compatibility and that the correct working directory is set so the server can resolve the project root for per-project state.