agent-network-protocol/anp

What it solves

AgentConnect provides a standardized way for AI agents to discover, authenticate, and communicate with one another across different programming languages. It solves the problem of fragmentation in the "agentic web" by providing a common protocol (ANP) for identity verification, interface discovery, and secure data exchange.

How it works

It implements the Agent Network Protocol (ANP) through a multi-language SDK (Python, Go, Rust, Dart, TypeScript, and Java). The system uses several core mechanisms:

• Identity & Discovery: Uses DID WBA (Web-based Decentralized Identifiers) for identity and Agent Description documents (ad.json) and OpenRPC/JSON-RPC for defining and calling agent interfaces.
• Security: Implements W3C Data Integrity proofs, HTTP Message Signatures, and end-to-end encryption (E2EE) for private and group communications.
• Naming: Uses WNS (WBA Name Space) to resolve human-readable handles to agent identities.
• OpenANP: A Python-specific framework that automatically turns Python methods into discoverable ANP interfaces using FastAPI.

Who it’s for

Developers building interoperable AI agents, services that need to verify agent identities, or platforms creating a network of autonomous agents that can securely call each other's tools.

Highlights

• Multi-language Support: SDKs available for Python, Go, Rust, Dart, TypeScript, and Java.
• Automatic Interface Generation: Python's OpenANP converts type hints directly into OpenRPC documents.
• Robust Security: Built-in support for E2EE (Direct and Group/MLS) and verifiable proofs.
• Decentralized Identity: Integration with DID WBA for secure, decentralized authentication.
• Human-Readable Handles: Support for WNS to make agents easier to find and address.