====== Agent Network Protocol ======
The **Agent Network Protocol (ANP)** is an open-source communication standard designed to enable secure, decentralized communication between AI agents across the open internet.((ANP GitHub Repository. [[https://github.com/agent-network-protocol/AgentNetworkProtocol|github.com/agent-network-protocol]])) Often described as "the HTTP of the Agentic Web era," ANP aims to build an open and efficient collaborative network for billions of agents while breaking down data silos and enabling autonomous agent organization. ANP is recognized as one of three key protocols documented for standardizing agent interaction in distributed multi-agent environments.((Cobus Greyling (LLMs) - Two-Thirds of Multi-Agent Intelligence (2026). [[https://cobusgreyling.substack.com/p/two-thirds-of-multi-agent-intelligence|cobusgreyling.substack.com]]))

===== Overview =====
Unlike centralized agent communication protocols that depend on platform-specific identity providers or API keys, ANP uses decentralized identifiers (DIDs) as the foundation for agent identity. This allows agents to autonomously control their identities without relying on any single authority, enabling permissionless agent discovery and networking across organizational boundaries.((Agent Network Protocol. Official Site. [[https://agent-network-protocol.com|agent-network-protocol.com]])) Inter-agent communication protocols represent a critical infrastructure layer in multi-agent systems, with communication often identified as a bottleneck rather than individual agent reasoning capability itself.(([[https://cobusgreyling.substack.com/p/two-thirds-of-multi-agent-intelligence|Cobus Greyling (LLMs) - Two-Thirds of Multi-Agent Intelligence (2026]]))

The protocol is developed and maintained by an open-source community that has pledged never to commercialize it, ensuring an open and neutral foundation for the agentic web.

===== Three-Layer Architecture =====
ANP employs a protocol stack with three distinct layers:

=== 1. Identity and Encrypted Communication Layer ===
Implements decentralized authentication and end-to-end encryption based on **W3C DID (Decentralized Identifier)** standards. Agents authenticate each other without centralized certificate authorities or identity providers. Notably, ANP does not depend on blockchain — it leverages existing web infrastructure (HTTPS, DNS) for deployment and resolution.((W3C Community Group. "AI Agent Protocol." [[https://w3c-cg.[[github|github]].io/ai-agent-protocol/|w3c-cg.[[github|github]].io]]))

=== 2. Meta-Protocol Layer ===
Solves dynamic protocol negotiation by allowing agents to automatically negotiate which protocol format, version, and communication pattern to use. This enables heterogeneous agents built on different frameworks to interoperate without pre-agreed specifications.

=== 3. Application Protocol Layer ===
Builds AI-friendly interfaces through the **Agent Description Protocol (ADP)** and discovery mechanisms. Agents expose their capabilities in structured, machine-readable formats that enable [[semantic_search|semantic search]] and automated service matching.

===== Authentication Flow =====
Agent-to-agent authentication in ANP follows a DID-based handshake:

  - Agent A initiates a connection request carrying its DID
  - Agent B resolves Agent A's DID document to obtain the public key
  - Agent B verifies the request signature using Agent A's public key
  - Agent B returns an access token
  - Agent A uses this token in all subsequent requests

This flow enables trustless, decentralized verification without any intermediary.((ANP Developer Guide. [[https://www.agent-network-protocol.com/guide/|agent-network-protocol.com/guide]]))

===== Decentralized vs Centralized Comparison =====
^ Aspect ^ ANP (Decentralized) ^ Centralized Protocols ^
| Identity Model | DID-based, self-sovereign | Centralized identity providers (OAuth/OIDC) |
| Data Silos | Broken via standardized protocols | Services remain isolated |
| Account Management | Single DID across all services | Separate accounts per platform |
| Dependencies | No centralized authority | Single points of failure |
| Agent Autonomy | Full self-governance | Limited by platform policies |

===== Code Example =====
Resolving an agent's DID document and establishing a connection:

<code python>
import requests
import json
from cryptography.hazmat.primitives.asymmetric import ed25519

AGENT_DID = "did:web:agent-b.example.com"
DID_RESOLUTION_URL = f"https://agent-b.example.com/.well-known/did.json"

Resolve the target agent's DID document
did_doc = requests.get(DID_RESOLUTION_URL).json()
public_key = did_doc["verificationMethod"][0]["publicKeyMultibase"]

Prepare signed connection request
my_private_key = ed25519.Ed25519PrivateKey.generate()
connection_request = {
    "from_did": "did:web:agent-a.example.com",
    "to_did": AGENT_DID,
    "purpose": "capability_query",
    "timestamp": "2026-03-24T12:00:00Z"
}

message = json.dumps(connection_request).encode()
signature = my_private_key.sign(message)

Send authenticated request
response = requests.post(
    f"https://agent-b.example.com/anp/connect",
    json=connection_request,
    headers={"X-ANP-Signature": signature.hex()}
)
access_token = response.json()["access_token"]
print(f"Connected to {AGENT_DID}, token: {access_token[:16]}...")
</code>

===== Core Challenges Addressed =====
ANP systematically addresses three fundamental problems in the agentic ecosystem:((arXiv. "Agent Network Protocol Specification." [[https://arxiv.org/abs/2508.00007|arXiv:2508.00007]]))

  * **Agent Interconnection** — Enabling all agents to connect regardless of developer, platform, or framework
  * **Data Silos and Context** — Breaking barriers so agents can access complete contextual information across services
  * **Autonomous Organization** — Using AI-driven negotiation and collaboration to form economically efficient agent networks

===== See Also =====
  * [[agent_communication_protocol|Agent Communication Protocol]]
  * [[agent_protocol|Agent Protocol]]
  * [[a2a_protocol|A2A Protocol]]
  * [[mcp_protocol|MCP Protocol]]
  * [[a2a_vs_mcp_vs_agui|A2A vs MCP vs AG-UI]]

===== References =====