Remote MCP Agent

This repository implements a Model Context Protocol (MCP) agent that can be deployed as a remote service. The agent architecture is designed to work with Cloudflare Workers and integrates with various AI models and services.

Project Structure

The project is organized into several key components:

• garden-agent: Implementation of the Garden Agent, which handles AI requests and responses
• mcp/animal: Animal-specific MCP implementation
• mcp/garden: Garden-specific MCP implementation
• a2a: Agent-to-Agent communication framework

Key Features

• Integration with Momento for caching and state management
• Support for various AI models from providers like Meta (Llama), Hugging Face, and others
• Cloudflare Workers-based deployment
• R2 storage integration for object storage
• WebSocket support for real-time communication

Technologies Used

• TypeScript
• Cloudflare Workers
• Momento Cache
• Various AI models including:
  • Meta Llama models (Llama 3, Llama 4 Scout)
  • Hugging Face models
  • Stability AI models
  • BAAI embedding models
  • And more

Getting Started

To run this project, you'll need:

1. Cloudflare Workers account
2. Momento API key
3. Anthropic API key

Environment Variables

The following environment variables are required:

• MOMENTO_API_KEY: API key for Momento cache
• ANTHROPIC_KEY: API key for Anthropic AI services

Deployment

The agent is designed to be deployed as a Cloudflare Worker. The main entry point is in garden-agent/src/index.ts.

Usage

The agent exposes an HTTP endpoint that can be called to interact with the MCP agent. The agent handles requests through the A2A (Agent-to-Agent) server implementation.

// Example of how the agent is initialized
const agentExecutor = new GardenAgent(anthropicApiKey);
const requestHandler = new MomentoRequestHandler(
  agentCard,
  agentExecutor,
  {
    cacheName: 'mcp',
    momentoApiKey,
    waitForAgentOnTaskCancellation: true,
  },
);

License

[Add license information here]