# AgentAnycast MCP Server

> MCP server for peer-to-peer AI agent networking. Connects any MCP-compatible AI tool to a decentralized network of AI agents with end-to-end encryption, NAT traversal, and skill-based routing.

## What it does

AgentAnycast MCP Server exposes 6 tools that let AI assistants (Claude, Cursor, VS Code, ChatGPT, Gemini CLI, etc.) discover and communicate with AI agents on a P2P network:

- `discover_agents(skill)` — Find agents by capability (e.g. "translate", "summarize")
- `send_task(target, message)` — Send encrypted task to agent (by PeerID, skill name, or URL)
- `get_task_status(task_id)` — Check task result
- `get_agent_card(peer_id)` — Get agent's capability card
- `list_connected_peers()` — List connected peers
- `get_node_info()` — Node identity and status

## Key features

- Zero config: `uvx agentanycast-mcp` — daemon auto-managed
- Zero API keys: Cryptographic identity (Ed25519 → did:key)
- End-to-end encrypted: Noise_XX through relays
- NAT traversal: Automatic hole-punching + relay fallback
- Skill-based routing: Send to "translate" not to an address
- Decentralized: No central server required

## Install

```
uvx agentanycast-mcp          # stdio mode
pip install agentanycast-mcp   # or install globally
```

## Links

- Source: https://github.com/AgentAnycast/agentanycast-mcp
- Main project: https://github.com/AgentAnycast/agentanycast
- Python SDK: https://github.com/AgentAnycast/agentanycast-python
- TypeScript SDK: https://github.com/AgentAnycast/agentanycast-ts
