agentanycast

Overview

AgentAnycast is a P2P runtime that enables AI agents to communicate across networks without requiring public IPs. It features end-to-end encryption, NAT traversal, and skill-based routing. The system includes an MCP server that allows integration with Claude Desktop, Cursor, VS Code, and other MCP-compatible AI tools. The architecture uses a sidecar model where a local Go daemon handles all P2P networking, encryption, and protocol logic, while thin SDKs in Python and TypeScript provide application interfaces.

Try asking AI

After installing, here are 4 things you can ask your AI assistant:

• find_agents

  Find agents by skill capability

  Try:Find agents that can translate JapaneseTry:Who can help me with data analysis?
• send_task

  Send a task to a specific agent or by skill

  Try:Send a task to the summarize agentTry:Use the translate skill to convert this text
• list_peers

  List all available peer agents

  Try:Show me all available agentsTry:List all connected peers

Note: Inferred from MCP server documentation and usage examples. The README mentions MCP tools for finding agents by skill and sending tasks, but doesn't provide explicit tool names or signatures.

Comparable tools

Installation

uvx agentanycast-mcp

{
  "mcpServers": {
    "agentanycast": {
      "command": "uvx",
      "args": ["agentanycast-mcp"]
    }
  }
}

pip install agentanycast

npm install agentanycast

FAQ

How does AgentAnycast work behind NATs and firewalls?

AgentAnycast uses DCUtR hole-punching with automatic NAT detection and falls back to encrypted relays when direct connection isn't possible.

What protocols does the MCP server support?

The MCP server supports both stdio and HTTP transport methods, mapping MCP Tools to A2A Skills for interoperability with AI tools.

Compare agentanycast with