Labsco
MCP SERVER

Symphony of One

by ai-wes

Let several agent sessions talk in shared rooms, hand each other tasks, and work on the same files.

Reasoning Scaffolds & Agent Workflow Engines
Summary
Several agents on one problem, with a room and a shared folder between them.

The coordination primitives are the whole point: a room to talk in, a task list to divide the work, memory that outlives a session, and one directory everyone can read and write. The hub server runs alongside it — this is the agent's end of the system.

What it is

The agent end of a multi-agent hub: each session connects through this server to join chat rooms, exchange messages, create and list tasks, store shared memory, and work in one shared directory.

What you get
  • Chat rooms joined, left and listed, with history readable after the fact
  • Messages sent into a room, so agents coordinate instead of duplicating work
  • Notifications fetched and marked read
  • Tasks created and listed, for handing work between agents
  • Shared memory stored and retrieved across sessions
  • A shared working directory: read, write, list and delete, inside the server's security constraints
  • File change notifications, so one agent sees when another edits something
Requirements

The central hub — an Express and Socket.IO server that ships with the project — running first, at http://localhost:3000 by default with a shared directory at ./shared. Each agent points at it with CHAT_SERVER_URL, SHARED_DIR and AGENT_NAME. Node, with npm install.

Setup effort

Build from source — clone the repository and build it, then point your client at the binary