Strands MCP Server

# Examples This directory contains example scripts demonstrating how to use `strands-mcp-server`. ## 📁 Files | File | Description | |------|-------------| | **mcp_server.py** | Start an MCP server exposing agent tools | | **mcp_test.py** | Connect to MCP server using mcp_client tool | ## 🚀 Quick Start ### 1. Start the MCP Server ```bash cd examples python mcp_server.py ``` **What it does:** - Creates agent with calculator and shell tools - Starts MCP server on port 8000 - Exposes tools via http://localhost:8000/mcp - Provides interactive agent prompt **Output:** ``` 🚀 Starting MCP Server Demo ============================================================ 📡 Starting MCP server on port 8000... ✅ MCP server 'demo-server' started on port 8000 📊 Mode: stateful (session persistence) 🔧 Exposed 3 tools: • calculator • shell • invoke_agent (full agent invocation) ✨ 🔗 Connect at: http://localhost:8000/mcp ============================================================ ✅ Server running! You can now: • Connect at: http://localhost:8000/mcp • Use with Claude Desktop (via npx mcp-remote) • Call tools remotely from other agents ============================================================ 💬 Agent is also available interactively! Try: 'show server status' or 'calculate 42 * 89' # ``` ### 2. Test with MCP Client (New Terminal) ```bash cd examples python mcp_test.py ``` **What it does:** - Connects to the MCP server - Lists available tools - Calls calculator tool remotely - Invokes full agent conversation - Demonstrates complete MCP workflow **Output:** ``` 🚀 MCP Client Test Demo ============================================================ 🔧 Method 1: Using mcp_client tool ------------------------------------------------------------ 📡 Connecting to MCP server at http://localhost:8000/mcp... ✅ Connected to MCP server: demo-server 📋 Listing available tools from server... 📊 Available tools from demo-server: • calculator - Calculator powered by SymPy • shell - Interactive shell with PTY support • invoke_agent - Full agent invocation ✨ 🧮 Calling remote calculator tool... Result: 887112 🤖 Invoking full agent remotely... [Agent response with calculator usage] 👋 Disconnecting... ✅ Disconnected from MCP server: demo-server ``` ## 🎓 Usage Patterns ### Pattern 1: Basic MCP Server ```python from strands import Agent from strands_tools import calculator, file_read from strands_mcp_server import mcp_server agent = Agent(tools=[calculator, file_read, mcp_server]) agent("start mcp server on port 8000") ``` ### Pattern 2: MCP Client Connection ```python from strands import Agent from strands_mcp_server import mcp_client agent = Agent(tools=[mcp_client]) agent("connect to http://localhost:8000/mcp as my-server") agent("list tools from my-server") agent("use calculator on my-server to compute 42 * 89") ``` ### Pattern 3: Agent-to-Agent Communication ```python # Server Agent from strands import Agent from strands_tools import calculator, shell from strands_mcp_server import mcp_server server_agent = Agent(tools=[calculator, shell, mcp_server]) server_agent("start mcp server") # Client Agent (different process/machine) from strands import Agent from strands_mcp_server import mcp_client client_agent = Agent(tools=[mcp_client]) client_agent("connect to http://server-host:8000/mcp as remote") client_agent("invoke full agent on remote: analyze system logs") ``` ## 🔌 Claude Desktop Integration Update `~/Library/Application Support/Claude/claude_desktop_config.json`: ```json { "mcpServers": { "my-strands-agent": { "command": "npx", "args": [ "-y", "mcp-remote", "http://localhost:8000/mcp", "--allow-http" ] } } } ``` Then restart Claude Desktop - your agent's tools will be available! ## 🛠️ Advanced Examples ### Expose Only Specific Tools ```python agent.tool.mcp_server( action="start", port=8000, tools=["calculator", "file_read"], # Only these tools expose_agent=False # Disable full agent invocation ) ``` ### Multiple Servers ```python # Public server (limited tools) agent.tool.mcp_server( action="start", server_id="public", port=8000, tools=["calculator", "current_time"] ) # Admin server (all tools) agent.tool.mcp_server( action="start", server_id="admin", port=8001 ) ``` ### Stateless Mode (Multi-Node) ```python # For load-balanced deployments agent.tool.mcp_server( action="start", port=8000, stateless=True # No session state ) ``` ## 📊 Monitoring Check server status: ```python agent.tool.mcp_server(action="status") ``` List all running servers: ```python agent.tool.mcp_server(action="list") ``` ## 🐛 Troubleshooting **Port already in use:** ```bash # Kill process using port 8000 lsof -ti:8000 | xargs kill -9 ``` **Server not responding:** ```bash # Check if server is running curl http://localhost:8000/mcp ``` **Import errors:** ```bash # Reinstall package pip install -e . ```