Claude Swarm MCP Server

A Model Context Protocol (MCP) server that enablesmulti-agent orchestrationusing Claude AI through Claude Desktop. Create, manage, and coordinate specialized AI agents for complex workflows like financial analysis, customer service, and research.

- 🤖 Persistent Agents: Create specialized Claude agents that survive restarts
- 🔄 Agent Coordination: Intelligent handoffs between agents based on expertise
- 💾 Local Storage: All agents and conversations saved locally
- 📊 Pre-built Templates: Ready-to-use financial analysis and customer service teams
- 🎯 Specialized Functions: Custom tools and capabilities per agent
- 🔧 Easy Integration: Works seamlessly with Claude Desktop

- Python 3.10+
- Claude Desktop installed
- Anthropic API key with billing enabled

git clone https://github.com/yourusername/claude-swarm-mcp.git cd claude-swarm-mcp

Edit~/Library/Application Support/Claude/claude_desktop_config.json:

{ "mcpServers": { "claude-swarm": { "command": "python3", "args": ["/path/to/claude-swarm-mcp/claude_swarm_mcp_server.py"], "env": { "ANTHROPIC_API_KEY": "your-api-key-here" } } } }

Create finance team with company name "Your Company"

Create finance team with company name "TechVest Capital"

- Risk Analyst- VaR calculations, stress testing
- Portfolio Manager- Asset allocation, optimization
- Data Analyst- Market data, performance metrics
- Research Analyst- Investment research, market analysis

Chat with agent: "Calculate the VaR for a portfolio with AAPL 30%, GOOGL 25%, MSFT 20%, AMZN 15%, TSLA 10%" using agent "Risk Analyst"

Create agent with name "Options Specialist" and instructions "You are an expert in options trading. Calculate Greeks, analyze volatility, and recommend hedging strategies."

Chat with agent: "I need a complete analysis of my tech portfolio: analyze risk, optimize allocation, and provide investment recommendations."

Agents will coordinate automatically to provide comprehensive analysis

claude-swarm-mcp/ ├── claude_swarm.py # Core Swarm framework ├── claude_swarm_mcp_server.py # MCP server implementation ├── requirements.txt # Python dependencies ├── README.md # This file ├── LICENSE # MIT License ├── examples/ # Usage examples │ ├── finance_workflow.py # Financial analysis example │ ├── customer_service.py # Customer service template │ └── research_team.py # Research coordination example ├── tests/ # Test suite │ ├── test_agents.py # Agent functionality tests │ ├── test_mcp_server.py # MCP server tests │ └── test_swarm.py # Swarm coordination tests └── docs/ # Documentation ├── API.md # API reference ├── DEPLOYMENT.md # Deployment guide └── CONTRIBUTING.md # Contribution guidelines

-

Claude Swarm Framework(claude_swarm.py)

- Multi-agent orchestration
- Automatic handoffs between agents
- Shared conversation context
- Function calling integration

MCP Server(claude_swarm_mcp_server.py)

- Model Context Protocol implementation
- Persistent agent storage
- Tool registration and handling
- Claude Desktop integration

- JSON-based agent persistence
- Conversation history
- Context variables
- Backup and restore capabilities

Claude Desktop ↔ MCP Protocol ↔ Swarm Server ↔ Claude API ↓ Agent Storage (JSON)

- Portfolio Risk Analysis: VaR calculations, stress testing
- Investment Research: Market analysis, stock recommendations
- Compliance Monitoring: Regulatory requirements, position limits
- Client Advisory: Personalized investment advice

- Intelligent Triage: Route customers to appropriate specialists
- Multi-language Support: Automatic language detection and routing
- Escalation Management: Seamless handoffs to senior agents
- Knowledge Base Integration: Context-aware information retrieval

- Literature Review: Coordinate research across multiple domains
- Data Analysis: Statistical analysis, visualization, reporting
- Project Management: Task coordination, milestone tracking
- Technical Documentation: Automated documentation generation

- Local Storage: All data stored locally on your machine
- API Key Security: Secure API key handling through environment variables
- No External Dependencies: No third-party services for agent storage
- Audit Trail: Complete conversation history and agent interactions

export ANTHROPIC_API_KEY="your-api-key" export CLAUDE_SWARM_STORAGE_DIR="/custom/storage/path" # Optional export CLAUDE_SWARM_DEBUG="true" # Optional debug mode

# Custom storage location storage_path = "/Users/yourname/claude_agents" server = ClaudeSwarmMCPServer(storage_dir=storage_path)

- Agent Creation: < 2 seconds
- Chat Response: 3-8 seconds (depending on complexity)
- Agent Handoffs: < 1 second
- Storage Operations: < 500ms
- Memory Usage: ~50-100MB (depending on conversation history)

# Check your API key python3 -c "from anthropic import Anthropic; print('API key valid')"

- Restart Claude Desktop
- Check server logs for errors
- Verify config file path and syntax

- Check billing status in Anthropic Console
- Verify agent instructions are clear
- Test with simple messages first

# Fix permissions chmod 755 /path/to/storage/directory

# Run server with debug logging CLAUDE_SWARM_DEBUG=true python3 claude_swarm_mcp_server.py

We welcome contributions! Please seeCONTRIBUTING.mdfor guidelines.

# Clone and setup development environment git clone https://github.com/yourusername/claude-swarm-mcp.git cd claude-swarm-mcp python3 -m venv venv source venv/bin/activate pip install -r requirements-dev.txt

This project is licensed under the MIT License - see theLICENSEfile for details.

- Issues:GitHub Issues
- Discussions:GitHub Discussions
- Documentation:docs/

git clone https://github.com/yourusername/claude-swarm-mcp.git cd claude-swarm-mcp

An MCP server for initial app planning and creating a good starting point for an app.