UTCP-MCP Bridge

# MCP Strands Agent Example ## Overview This example demonstrates how to build an AI agent using the **Strands SDK** that connects to **UTCP-MCP Bridge** servers to access and utilize tools through the Model Context Protocol (MCP). The agent provides a conversational interface where users can interact with various tools exposed by the UTCP ecosystem. ## What This Example Does The MCP Strands Agent (`mcp_strands_agent.py`) creates an intelligent assistant that: - **Connects to UTCP-MCP Bridge servers** to discover available tools - **Uses Strands SDK** to create a conversational AI agent powered by Amazon Bedrock's Claude 3 Sonnet - **Maintains conversation context** across multiple interactions - **Provides a command-line interface** for natural language tool interaction - **Handles multiple MCP servers** simultaneously for maximum tool availability ## Key Features ### UTCP-MCP Integration - Connects to the UTCP Proxy MCP server (port 8777) by default - Automatically discovers and loads all available tools from connected servers - Seamlessly bridges UTCP tools to the Strands agent framework ### Conversational Interface - Natural language interaction with AI-powered tool selection - Persistent conversation history within a session - Built-in commands for session management (`/help`, `/clear`, `/quit`) ### Error Handling & Resilience - Graceful handling of server connection failures - Continues operation even if some MCP servers are unavailable - Detailed error reporting and debugging information ## Prerequisites Before running this example, ensure you have: 1. **UTCP-MCP Bridge running** (see main project README) 2. **AWS credentials configured** with Amazon Bedrock access for Claude 3 Sonnet 3. **Python 3.8+** installed 4. **Required Python packages** (see requirements.txt) ## Quick Start ### 1. Start UTCP-MCP Bridge From the main project directory: ```bash # Using Docker docker-compose up --build # OR using local script ./run.sh ``` This will start the UTCP-MCP Bridge servers on: - Port 8777 (UTCP Proxy MCP) - Port 8776 (UTCP Client MCP) - Port 8778 (Web UI) ### 2. Install Dependencies ```bash pip install -r requirements.txt ``` ### 3. Configure AWS Credentials Ensure your AWS credentials are configured with Amazon Bedrock access for Claude models: ```bash aws configure # OR set environment variables export AWS_ACCESS_KEY_ID=your_key export AWS_SECRET_ACCESS_KEY=your_secret export AWS_REGION=us-east-1 ``` **Note:** Your AWS account must have access to Amazon Bedrock and the Claude 3 Sonnet model in the specified region. ### 4. Run the Agent ```bash python mcp_strands_agent.py ``` ## Usage Once the agent starts, you'll see: ``` Initializing MCP clients... Connecting to http://localhost:8777/utcp-proxy... Found X tools from http://localhost:8777/utcp-proxy Total available tools: X - Name: tool1; Type: function - Name: tool2; Type: function ... MCP agent initialized. You can now start your conversation. Type /help to see available commands. Enter your query (I can use available MCP tools to help). /help: ``` ### Available Commands - `/help` - Show available commands - `/clear` - Clear conversation history and start fresh - `/quit`, `/bye`, `/exit` - End the session ### Example Interactions ``` Enter your query: What tools do you have available? Response: I have access to several tools through the UTCP-MCP Bridge... Enter your query: Can you help me search for files in my project? Response: I can help you search for files. Let me use the file search tool... ``` ## Configuration ### MCP Server URLs Modify the `MCP_SERVERS` list in the script to connect to different servers: ```python MCP_SERVERS = [ "http://localhost:8777/utcp-proxy", # UTCP Proxy MCP "http://localhost:8776/utcp-client", # UTCP Client MCP (optional) # Add more MCP servers as needed ] ``` ### AI Model Change the Amazon Bedrock model by modifying the `MODEL_ID` variable: ```python MODEL_ID = "anthropic.claude-3-sonnet-20240229-v1:0" # Default - Claude 3 Sonnet # MODEL_ID = "anthropic.claude-3-haiku-20240307-v1:0" # Claude 3 Haiku - Faster, less capable ``` **Note:** Ensure the selected model is available in your AWS region and that your account has access to it through Amazon Bedrock. ### System Prompt Customize the agent's behavior by modifying the `system_prompt` variable to change how the AI assistant behaves and responds to user queries. ## Architecture ``` ┌─────────────────┐ ┌──────────────────┐ ┌─────────────────────┐ │ User Input │───▶│ Strands Agent │───▶│ Amazon Bedrock │ └─────────────────┘ └──────────────────┘ │ (Claude Models) │ │ └─────────────────────┘ ▼ ┌──────────────────┐ │ MCP Clients │ └──────────────────┘ │ ▼ ┌──────────────────┐ │ UTCP-MCP Bridge │ └──────────────────┘ │ ▼ ┌──────────────────┐ │ UTCP Tools │ └──────────────────┘ ``` ## Troubleshooting ### Common Issues 1. **"No MCP clients could be created"** - Ensure UTCP-MCP Bridge is running - Check that ports 8777/8776 are accessible - Verify server URLs in `MCP_SERVERS` 2. **AWS/Amazon Bedrock errors** - Verify AWS credentials are configured - Check AWS region is set to `us-east-1` (or your preferred Bedrock region) - Ensure you have access to Amazon Bedrock and Claude models - Verify the model ID exists and is available in your region 3. **"No tools available"** - Check UTCP-MCP Bridge web UI at http://localhost:8778 - Verify UTCP providers are registered and tools are enabled - Check server logs for connection issues ### Debug Mode For detailed debugging, the script includes comprehensive error handling and traceback printing. Monitor the console output for specific error messages. ## Integration with UTCP Ecosystem This example showcases the power of the UTCP-MCP Bridge by: - **Leveraging UTCP's universal tool protocol** through MCP compatibility - **Accessing any UTCP-registered tools** without direct UTCP client integration - **Demonstrating seamless tool discovery** and usage in AI applications - **Providing a template** for building more complex UTCP-powered agents ## Next Steps - Explore the UTCP-MCP Bridge web UI at http://localhost:8778 - Register additional UTCP providers to expand tool availability - Customize the agent's system prompt for specific use cases - Build more sophisticated workflows using the discovered tools ## Related - [UTCP-MCP Bridge Main README](../../README.md) - [Strands SDK Documentation](https://github.com/strands-ai/strands) - [Model Context Protocol Specification](https://modelcontextprotocol.io/)