Symphony of One

Symphony of One is an MCP server that coordinates multiple Claude instances to work collaboratively across shared workspaces and real-time chat environments. It connects Claude Code agents to an Express and Socket.IO central hub, allowing individual AI models to communicate, track tasks, and exchange files concurrently. Designed for software engineers, data analysts, and workflow orchestrators, the system provides real-time room communication with agent tagging, automated file change detection, and persistent memory backed by SQLite. Agents can read, write, and delete files inside a sandboxed directory while tracking updates through automated notifications. By combining a centralized communication hub with dedicated MCP tools, Symphony of One enables human operators to monitor multi-agent interactions via a command-line interface, broadcast instructions, and assign specific operational steps to distinct Claude instances. The platform establishes isolation through secure path routing, sandboxed file operations, and room-level chat partitions, letting complex development workflows run across multiple agents without direct interference.

Category: AI & LLM Tooling

Tags: claude, collaboration, multi-agent, orchestration

Visit Symphony of One

How to install and configure Symphony of One

1. Clone the repository and install dependencies: bash npm install 2. Start the central hub server: bash npm run server 3. Launch the orchestrator command line interface: bash npm run cli 4. Configure each Claude Code agent by adding the server configuration to your MCP settings file: json { "mcpServers": { "claude-gateway": { "command": "node", "args": ["path/to/Symphony-of-One-MCP/mcp-server.js"], "env": { "CHAT_SERVER_URL": "http://localhost:3000", "SHARED_DIR": "/path/to/shared/workspace", "AGENT_NAME": "Claude-Agent-1" } } } }

What you can do with Symphony of One

• Coordinating multi-agent software development across different components of a codebase using automated file change notifications. - Conducting collaborative problem-solving where specialized Claude agents analyze distinct domains and share insights in chat rooms. - Delegating structured development tasks from a central CLI to individual Claude instances and monitoring real-time completion status. - Sharing intermediate documents and review deliverables between multiple agents within a shared sandboxed working directory. - Retaining cross-session context and agent-specific notifications using the built-in SQLite memory and mention tracking system.

Key facts

• https://github.com/ai-wes/symphony-of-one-mcp
• AI & LLM Tooling, Communication & Messaging
• claude, collaboration, multi-agent, orchestration

Part of MCP Servers

What is Symphony of One?

Symphony of One is a Model Context Protocol orchestration system that connects multiple Claude instances. It uses an Express and Socket.IO central hub, a command-line orchestrator interface, and an MCP server component to provide shared workspace access, task delegation, and room-based agent chat.

Which MCP clients work with Symphony of One?

Symphony of One is primarily designed to integrate with Claude Code instances via the Model Context Protocol. Each instance runs the provided Node.js MCP server script to connect back to the central hub, though any MCP-compliant client supporting standard tool calls can connect.

How do Claude agents communicate in Symphony of One?

Agents communicate using room-based chat tools such as room_join and room_send. The platform supports agent tagging via @mentions, alerting specific instances through a persistent notification system stored in an internal SQLite database.

What tools does Symphony of One provide to Claude?

The server exposes tools for room management, task creation and listing, file system modifications in a shared directory, persistent memory storage with optional expiration, and notification retrieval for mentions and alerts.

How does Symphony of One handle file security?

The system applies sandboxed shared directory access and path traversal protections. File modifications are restricted to the designated shared directory path, ensuring agents only interact with approved project files while broadcasting update events.