Skynet-MCP (THIS PROJECT IS A WORK IN PROGRESS)

A hierarchical network of AI agents using the Model Context Protocol (MCP). The AI Agents in this network can spawn new agents to have them do work, each agent also includes all the tools that the initiating agent has. The network of agents are capable of working through complex tasks, ranging from research, reporting, to coding.

By leveraging the growing collection of MCP-servers available each of the agents potentially have a large set of tools at their disposal to do all kinds of tasks.

Want to help?

If you like this idea and are proficient in TypeScript; I can use your help! Please reach out if you like to contribute.

Overview

Skynet-MCP is an advanced architecture that implements the Model Context Protocol (MCP) to create a hierarchical network of AI agents. Each Skynet-MCP instance acts as both an MCP server and MCP client, allowing it to provide tools to parent agents while also spawning and managing child agents.

This architecture enables recursive agent networks capable of decomposing complex tasks, parallelizing work, and integrating specialized capabilities across multiple models and services.

Features

• Dual-Mode Operation: Acts as both MCP server and client
• LLM Integration: Connect to OpenAI or Anthropic models for intelligent decision-making
• Tool Discovery: Automatically discover and use tools from connected MCP servers
• Hierarchical Agent Management: Create and manage child agents
• Configurable: Flexible configuration system supporting multiple environments
• FastMCP Integration: Built on the efficient FastMCP framework for robust MCP server implementations
• Multiple Transport Options: Supports both STDIO and SSE (Server-Sent Events) transport
• Delayed Execution: Run tasks asynchronously and poll for results later

MCP Tools

Skynet-MCP provides the following MCP tools through its FastMCP implementation:

1. Invoke: Creates and manages agent tasks

   • Parameters:
     • mcpConfig: MCP server configuration (available tools for the agent)
     • llmConfig: LLM configuration (provider and model to use)
     • prompt: Instructions for the agent task
     • delayedExecution: Boolean indicating if the task should be run asynchronously

2. DelayedResponse: Retrieves the result of asynchronous tasks

   • Parameters:
     • taskId: The ID of the task to check
   • Returns: Task status and result (if completed)

Agent

The agent framework still needs to be implemented.

Project Structure

skynet-mcp/
├── src/
│   ├── server/       # MCP server implementation
│   ├── client/       # MCP client implementation
│   ├── orchestrator/ # Agent orchestration
│   ├── tools/        # Agent tools
│   ├── persistence/  # State management
│   └── utils/        # Shared utilities
├── test/
│   ├── unit/         # Unit tests
│   └── integration/  # Integration tests
├── examples/         # Example implementations