AgentAnycast
Discover and communicate with AI agents over encrypted P2P networks. Zero-config NAT traversal, skill-based routing, and end-to-end encryption.
README
AgentAnycast Node
P2P daemon for decentralized A2A agent-to-agent communication.
AgentAnycast is fully decentralized. On a local network, it works with zero configuration via mDNS auto-discovery. For cross-network communication, just deploy your own relay with a single command.
Overview
AgentAnycast Node (agentanycastd) is the core daemon that powers the AgentAnycast network. It runs on each machine and handles:
- Automatic peer discovery via mDNS on local networks
- NAT traversal via circuit relay, hole punching, and QUIC
- End-to-end encryption using Noise_XX (Curve25519 + ChaCha20-Poly1305)
- A2A task routing with direct, skill-based, and HTTP bridge addressing
- Streaming for chunked artifact delivery
- HTTP Bridge for P2P ↔ HTTP A2A interop
- MCP Server for AI tool integration (Claude, Cursor, ChatGPT, etc.)
- ANP Bridge for Agent Network Protocol interop
- Prometheus metrics for observability
- gRPC API for language SDKs (Python, TypeScript) to interact with the daemon
Quick Start
Local network -- zero configuration
# Build
go build -o agentanycastd ./cmd/agentanycastd
# Run -- agents on the same LAN discover each other automatically
./agentanycastd
Cross-network -- deploy your own relay
# On any VPS with a public IP
git clone https://github.com/AgentAnycast/agentanycast-relay && cd agentanycast-relay
docker-compose up -d
# Note the RELAY_ADDR from the logs, then:
./agentanycastd -bootstrap-peers "/ip4/<RELAY_IP>/tcp/4001/p2p/12D3KooW..."
MCP mode -- use as an AI tool
# stdio mode (Claude Desktop, Cursor, VS Code, Gemini CLI)
./agentanycastd -mcp
# Streamable HTTP mode (ChatGPT, remote clients)
./agentanycastd -mcp-listen :3000
Configuration
Priority: CLI flags > environment variables > config file > defaults
CLI Flags
| Flag | Description |
|---|---|
-key |
Path to identity key file |
-grpc-listen |
gRPC listen address (unix:// or tcp://) |
-log-level |
Log level (debug, info, warn, error) |
-bootstrap-peers |
Comma-separated bootstrap multiaddrs |
-bridge-listen |
HTTP bridge listen address (e.g., :8080) |
-enable-webtransport |
Enable WebTransport (QUIC-based, browser-compatible) |
-mcp |
Run as MCP server over stdio |
-mcp-listen |
MCP Streamable HTTP listen address (e.g., :3000) |
-anp-listen |
ANP bridge listen address (e.g., :8090) |
-config |
Path to TOML config file |
-version |
Print version and exit |
Environment Variables
| Variable | Description | Default |
|---|---|---|
AGENTANYCAST_KEY_PATH |
Path to the libp2p identity key file | ~/.agentanycast/key |
AGENTANYCAST_GRPC_LISTEN |
gRPC server listen address | unix://~/.agentanycast/daemon.sock |
AGENTANYCAST_LOG_LEVEL |
Log level | info |
AGENTANYCAST_STORE_PATH |
Path to persistent data store | ~/.agentanycast/data |
AGENTANYCAST_BOOTSTRAP_PEERS |
Comma-separated relay/bootstrap multiaddrs | (none) |
AGENTANYCAST_ENABLE_MDNS |
Enable mDNS local network discovery | true |
AGENTANYCAST_REGISTRY_ADDRS |
Comma-separated registry addresses (multi-relay federation) | (none) |
AGENTANYCAST_MCP_LISTEN |
MCP Streamable HTTP address | (none) |
Config File
Default location: ~/.agentanycast/config.toml
key_path = "~/.agentanycast/key"
grpc_listen = "unix://~/.agentanycast/daemon.sock"
log_level = "info"
log_format = "json"
store_path = "~/.agentanycast/data"
enable_mdns = true
enable_quic = true
enable_webtransport = false
enable_relay_client = true
enable_hole_punching = true
offline_queue_ttl = "24h"
bootstrap_peers = ["/ip4/203.0.113.50/tcp/4001/p2p/12D3KooW..."]
[bridge]
enabled = false
listen = ":8080"
# tls_cert = "/path/to/cert.pem"
# tls_key = "/path/to/key.pem"
# cors_origins = ["*"]
[anycast]
routing_strategy = "random"
cache_ttl = "30s"
auto_register = true
# registry_addr = "relay.example.com:50052"
# registry_addrs = ["relay1:50052", "relay2:50052"] # federation
enable_dht = false
dht_mode = "auto" # "auto", "server", or "client"
[metrics]
enabled = false
listen = ":9090"
[mcp]
enabled = false
listen = ":3000"
[anp]
enabled = false
listen = ":8090"
[identity]
# did_web = "did:web:example.com:agents:myagent"
# did_dns_domain = "example.com"
Architecture
┌──────────────────────────────────────────────────────────────────┐
│ agentanycastd │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────────────┐ │
│ │ Engine │ │ Router │ │ Offline │ │ Anycast Router │ │
│ │(task FSM) │ │(A2A msg) │ │ Queue │ │(skill discovery) │ │
│ └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────────────┘ │
│ │ │ │ │ │
│ ┌────┴──────────────┴─────────────┴──────────────┴─────────┐ │
│ │ libp2p Host │ │
│ │ mDNS · Noise · TCP · QUIC · WebTransport · DHT │ │
│ │ Circuit Relay v2 · DCUtR Hole Punching │ │
│ └──────────────────────┬───────────────────────────────────┘ │
│ │ │
│ ┌──────────────────────┴──────────────────────────────────────┐ │
│ │ gRPC Server (16 RPCs for SDKs) │ │
│ └─────────────────────────────────────────────────────────────┘ │
│ │
│ ┌────────────┐ ┌────────────┐ ┌──────────┐ ┌────────┐ ┌──────┐ │
│ │ HTTP Bridge│ │ MCP Server │ │ANP Bridge│ │Metrics │ │BoltDB│ │
│ │ (A2A↔P2P) │ │(stdio/HTTP)│ │(ANP↔A2A) │ │(Prom.) │ │(stor)│ │
│ └────────────┘ └────────────┘ └──────────┘ └────────┘ └──────┘ │
└──────────────────────────────────────────────────────────────────┘
Internal Packages
| Package | Responsibility |
|---|---|
internal/a2a/ |
A2A protocol engine — task state machine, envelope routing, offline queue, streaming |
internal/node/ |
libp2p host — peer connections, mDNS, DHT, TCP/QUIC/WebTransport |
internal/crypto/ |
Ed25519 key management, DID conversion (did:key, did:web, did:dns), Verifiable Credentials |
internal/nat/ |
AutoNAT detection, DCUtR hole punching, Circuit Relay v2 client |
internal/store/ |
BoltDB persistence — tasks, agent cards, offline message queue |
internal/config/ |
Configuration — TOML file, environment variables, CLI flags |
internal/bridge/ |
HTTP Bridge — translates HTTP JSON-RPC ↔ P2P A2A envelopes |
internal/anycast/ |
Anycast router — skill-based addressing, registry + multi-registry federation + DHT discovery |
internal/metrics/ |
Prometheus metrics — connections, tasks, routing, bridge, streaming, MCP |
internal/mcp/ |
MCP Server — exposes P2P capabilities as MCP tools (stdio + Streamable HTTP) |
internal/anp/ |
ANP Bridge — translates ANP HTTP ↔ A2A P2P (JSON-RPC 2.0 + JSON-LD) |
pkg/grpcserver/ |
gRPC server — 16 RPC methods for SDKs |
gRPC API (16 RPCs)
| Group | Methods |
|---|---|
| Node | GetNodeInfo, SetAgentCard |
| Peers | ConnectPeer, ListPeers, GetPeerCard |
| Task Client | SendTask (peer_id / skill_id / url), GetTask, CancelTask, SubscribeTaskUpdates |
| Task Server | SubscribeIncomingTasks, UpdateTaskStatus, CompleteTask, FailTask |
| Streaming | SubscribeTaskStream, SendStreamingArtifact |
| Discovery | Discover |
MCP Server
The daemon can run as an MCP (Model Context Protocol) server, exposing P2P capabilities as tools for AI assistants:
| Tool | Description |
|---|---|
toolGetNodeInfo |
Get local node information |
toolListConnectedPeers |
List connected peers |
toolGetAgentCard |
Get local or remote agent card |
toolDiscoverAgents |
Discover agents by skill |
toolSendTask |
Send task to peer by ID |
toolSendTaskBySkill |
Send task by skill (anycast) |
toolGetTaskStatus |
Get task status |
Two transport modes:
- stdio — for local AI tool integration (Claude Desktop, Cursor, VS Code, Gemini CLI, JetBrains)
- Streamable HTTP — for remote clients (ChatGPT)
HTTP Bridge
Exposes an A2A-compatible HTTP endpoint for P2P ↔ HTTP interop:
GET /.well-known/a2a-agent-card— Agent Card discoveryPOST /— JSON-RPC endpoint for task operations- Optional TLS and CORS support
ANP Bridge
Exposes an ANP-compatible HTTP endpoint for Agent Network Protocol interop:
GET /agent/ad.json— Agent Description (JSON-LD)GET /agent/interface.json— OpenRPC specificationPOST /agent/rpc— JSON-RPC 2.0 endpoint
Metrics
Prometheus metrics on configurable HTTP port (default :9090):
agentanycast_connected_peers— current peer count (gauge)agentanycast_connections_total— connection events by direction (counter)agentanycast_connections_by_transport— connections by transport type: tcp/quic/webtransport (counter)agentanycast_tasks_total— tasks by direction and status (counter)agentanycast_task_duration_seconds— task latency histogramagentanycast_route_resolutions_total— anycast resolution by result (counter)agentanycast_bridge_requests_total— HTTP bridge requests (counter)agentanycast_stream_chunks_total— streaming chunks by direction (counter)agentanycast_messages_total— A2A messages by envelope type (counter)agentanycast_offline_queue_size— queued offline messages (gauge)agentanycast_mcp_tool_calls_total— MCP tool calls by tool and status (counter)
Disclaimer
This software is provided "as is", without warranty of any kind. This software uses cryptography and may be subject to export controls in certain jurisdictions.
License
FSL-1.1-ALv2 -- Functional Source License, Version 1.1, with Apache License, Version 2.0 as the future license. Each release converts to Apache 2.0 two years after its publication date.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.