Multi-Agent Discovery Protocol

Version 1.0.0 · Updated 2026-06-12 · ← Back to Specification

The Multi-Agent Discovery Protocol (MADP) defines how AI agents discover, evaluate, and coordinate with other agents across distributed web infrastructure. It extends Agent-First SEO principles to agent-to-agent (A2A) communication surfaces.

1. Problem Statement

As agentic workflows decompose tasks across specialized agents, agents need a standard mechanism to locate peer agents, understand their capabilities, and establish trust before delegating. Without a discovery protocol, agents resort to hardcoded endpoints or centralized registries — both fragility points.

2. Discovery Surface

An agent-capable domain exposes discovery at two layers:

Passive: /.well-known/agent.json — capabilities manifest readable without authentication
Active: /.well-known/agent-discovery.json — full entry-point map including feeds, constellation links, and trust anchors

3. Capability Declaration

{"protocol":"agent-seo/1.0","agent_type":"intelligence_hub","capabilities":["search","summarize","classify"],"input_formats":["text/plain","application/json"],"output_formats":["application/json","text/markdown"],"trust_anchor":"https://agent-seo.org/.well-known/trust-resolver.json","rate_limits":{"requests_per_minute":60}}

4. Constellation Membership

Agents that participate in a constellation expose their membership in /.well-known/constellation.json. This allows a discovering agent to traverse the full mesh and understand which domains share a trust lineage.

5. Trust Verification Flow

Discovering agent fetches /.well-known/agent.json
Reads trust_anchor URL and verifies trust-resolver chain
Checks constellation membership if multi-hop coordination needed
Proceeds with capability-scoped delegation

6. Conformance

Level A: Expose /.well-known/agent.json with protocol and capabilities fields.
Level AA: Add trust_anchor and rate_limits.
Level AAA: Full agent-discovery.json with constellation membership and input/output format declarations.