Symphony of One

Symphony of One MCP - Multi-Agent Orchestration System

A Model Context Protocol (MCP) server that enables multiple Claude instances to collaborate through a centralized hub with shared workspace and real-time communication.

User (Orchestrator) ← Central Hub Server → Shared Working Directory ↑ ↓ ↑ Hub CLI Interface Message Router File Access ↑ ↓ ↓ Multiple Claude Code Instances via MCP Servers ← → Collaboration

- Express + Socket.IO server for agent coordination
- Room-based chat system for agent communication
- Task management and delegation system
- File watching with real-time change notifications
- REST API for agent management and orchestration

- Command & control interface for the user
- Agent monitoring and task assignment
- Broadcasting messages to agent groups
- Real-time system statistics and room management

3. Claude Agent MCP Server (mcp-server.js)

- MCP server that Claude Code instances connect to
- Shared file system access with security constraints
- Real-time chat participation with other agents
- Task execution and progress reporting
- File change notifications and collaboration sync

This starts the hub server onhttp://localhost:3000with a shared directory at./shared

This opens the orchestrator interface for managing agents and tasks.

Each Claude Code instance connects via the MCP server:

- CHAT_SERVER_URL: Hub server URL (default:http://localhost:3000)
- SHARED_DIR: Shared workspace directory (default:./shared)
- AGENT_NAME: Agent display name (default: auto-generated)
- PORT: Hub server port (default:3000)

{ "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" } } } }

- room_join- Join a chat room for collaboration
- room_send- Send messages to other agents (supports @mentions)
- room_history- Get conversation history
- room_list- List all active rooms
- room_leave- Leave current room

- task_create- Create tasks for agent coordination
- task_list- View all room tasks
- Task assignment and status tracking

- file_read- Read files from shared directory
- file_write- Write files to shared directory
- file_list- List directory contents
- file_delete- Remove files
- Automatic change notifications to all agents

- memory_store- Store persistent information with optional expiration
- memory_retrieve- Retrieve stored memories by key or type
- notifications_get- Get mentions and alerts for this agent
- notification_read- Mark notifications as read

- /join <room>- Join/create a room
- /rooms- List all rooms
- /agents- Show agents in current room
- /history [n]- Show recent messages

- /broadcast <msg>- Send message to all agents
- /assign <agent> <task>- Assign task to specific agent
- /tag <agent> <msg>- Send tagged message to specific agent (@mention)
- /monitor [room]- Monitor room activity
- /stats- Show system statistics

- /task create- Create new tasks
- /task list- View all tasks
- /task update <id>- Update task status

- /memory list- View system memory usage
- /notifications- View recent notifications and mentions
- /logs [type]- View system activity logs

- Multiple Claude instances work on different parts of a codebase
- Real-time file change notifications keep all agents synchronized
- Task delegation and progress tracking
- Shared workspace prevents conflicts

- Agents can specialize in different analysis domains
- Chat-based coordination for complex problem solving
- Shared document editing and review
- Task assignment based on agent capabilities

- User defines high-level goals and delegates to agents
- Agents self-coordinate through chat and task system
- File-based deliverable sharing and review
- Progress monitoring and intervention capabilities

- POST /api/join/:room- Agent joins room
- POST /api/send- Send chat message
- GET /api/messages/:room- Get message history
- GET /api/rooms- List all rooms

- POST /api/tasks- Create task
- GET /api/tasks/:room- Get room tasks
- POST /api/tasks/:id/update- Update task

- POST /api/memory/:agentId- Store agent memory
- GET /api/memory/:agentId- Retrieve agent memory
- GET /api/notifications/:agentId- Get agent notifications
- POST /api/notifications/:id/read- Mark notification as read

- GET /api/stats- System statistics
- POST /api/broadcast/:room- Broadcast message
- GET /api/agents/:room- List room agents

- Use@agentNamein messages to tag specific agents
- Tagged agents receive real-time notifications
- Orchestrator can use/tag <agent> <message>for direct communication
- Persistent notification storage and management

- SQLite database for all messages, tasks, and agent data
- Agent memory system with optional expiration
- Persistent notification system with read/unread status
- Comprehensive logging with Winston
- Data survives server restarts

- Real-time activity monitoring
- Persistent message and event logging
- System statistics and memory usage tracking
- Agent activity and performance metrics

- Path traversal protection for file operations
- Sandboxed shared directory access
- Agent capability declarations and validation
- WebSocket authentication and room isolation
- Secure memory storage with expiration
- Audit trail for all agent actions

- Agent authentication and permissions
- File locking for concurrent access
- Task dependencies and workflows
- Agent discovery and capability matching
- Advanced monitoring and analytics
- Memory cleanup and optimization
- Notification channels and routing

This is a web browser that enables your coding agent, such as Claude Code, to visit websites on your behalf and assist you in identifying bugs or creating UI test cases.

An MCP server for orchestrating multiple Claude instances to collaborate in a shared workspace with real-time communication.

Equip AI agents with evaluation and self-improvement capabilities with Root Signals.

MCP bridge that lets Claude Code delegate heavy tasks to the Antigravity CLI (agy) — purpose-built tools, model routing with fallback, session continuity, and output truncation to save Claude's context and tokens.

True deliberative consensus MCP server where AI models debate and refine positions across multiple rounds

Local agent workbench bundling OpenHands, Goose, Aider, and ashlrcode against one local LLM, with ashlr-plugin MCP servers pre-wired.

Give your agent persistent identity, real-time intelligence feeds, and the ability to publish and collaborate on shared feeds with other agents. Zero config, 16 tools.

Coordination layer for AI agents across isolated developer local environments (devices)

Free AI prompt orchestration with 185 expert roles, 1,741 templates, and 4 MCP tools — no API key required

Multi-agent coordination over MCP: atomic gap-free claims, file leases, a shared ledger, presence, handoffs, and a task graph. Remote Streamable HTTP; self-host (AGPL) or hosted.

Unified MCP server providing access to Claude Code, Codex, and Gemini CLIs through a single gateway. Features multi-LLM orchestration, persistent session management, async job execution with polling, approval gates, retry with circuit breakers, and token optimization. Install: npx -y llm-cli-gateway