What it does
AgentAnycast Node runs as a sidecar daemon on each machine and provides P2P networking, end-to-end encryption and A2A protocol routing. Python and TypeScript SDKs communicate with it over gRPC. In MCP mode it exposes P2P capabilities as tools for AI assistants: node information, connected peers, agent cards, discovery by skill, sending tasks by peer ID or skill, and task status. An MCP Remote Proxy can also wrap an existing MCP server as a P2P-accessible agent.
It is relevant when agents on separate machines need to find and message each other. On a local network, agents discover one another through mDNS without configuration. For communication across networks, it uses NAT traversal and Circuit Relay v2. Security features include NaCl box encryption, DID identities (did:key, did:web, did:dns), skill-based access control and per-peer rate limiting. Interoperability features include an HTTP bridge, an ANP bridge, audit logging, Prometheus metrics and OpenTelemetry tracing.
The daemon is configured through CLI flags, environment variables or a TOML config file, with flags taking priority over the other two. Cross-network use requires deploying your own relay. MCP mode runs over stdio for local tools such as Claude Desktop, Cursor, VS Code, Gemini CLI and JetBrains, or over Streamable HTTP for remote clients such as ChatGPT.
Limitations and requirements
Cross-network communication requires deploying your own relay. The daemon runs on each machine that takes part in the network.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
AgentAnycast
P2P runtime for A2A protocol. Encrypted agent communication with zero-config NAT traversal.
v0.3.2As shown on registry.modelcontextprotocol.io when we last read it