Agent Discovery Exchange (AX)
Defining the social contract by which agents introduce themselves to the internet.
As AI agents move from isolated tools to distributed, autonomous systems, a new problem is becoming impossible to ignore:
How do agents discover each other, understand capabilities, and interoperate across organizational boundaries?
Today, there is no internet-native answer.
Integrations are hard-coded. Capabilities are documented out-of-band. Federation is bespoke. Governance is fragile.
That’s the gap Agent Discovery Exchange (AX) is intended to address.
The Missing Layer in Agent Systems
We already have:
- DNS for locating services
- HTTPS for exchanging structured information
- Mature and nascent**** execution protocols (REST, GraphQL, MCP, A2A)
What we don’t have is a standard way for an agent to say:
- Who I am
- What I can do
- How you can talk to me
- Under what constraints and trust assumptions
AX introduces that missing discovery and capability exchange layer, without redefining how agents execute.
What AX Is (and Isn’t)
AX is:
- A lightweight, protocol-agnostic discovery mechanism
- Built on existing internet standards
- Decentralized by design
- Execution-neutral
AX is not:
- A new agent protocol
- A registry
- A governance authority
- A replacement for MCP, A2A, REST, or GraphQL
AX exists to enable those protocols to interoperate dynamically.
How It Works (Conceptually)
At a high level:
- An agent publishes a small, machine-readable document at a well-known HTTPS location
- That document describes:
- Supported interaction protocols
- High-level capabilities
- Optional constraints and trust signals
- Other agents or supervisors and arbiters retrieve the document
- Local policy decides if and how to interact
No central registry. No global authority. No mandatory semantics.
Status and Next Steps
AX is early, evolving, and intentionally scoped.
It is published as Internet-Draft–style documents to:
- Encourage discussion
- Enable experimentation
- Invite independent implementation
Feedback, critique, and alternative designs are welcome.
Why Multiple Drafts?
AX is being developed as a series of Internet-Drafts, each layering additional capability, starting with core discovery, then planning for:
- Operational stability
- Trust and constraints
- Federation and feedback
This is intentional.
Discovery must stand on its own before trust, governance, and federation are layered on top. Each draft is independently useful and open to critique.
Learn More
- GitHub repository: https://github.com/sempfa/agent-discovery-rfc
- First draft: draft-agent-discovery-00.md
- Issues and discussions : open
Final Thought
Agentic AI systems are becoming distributed systems of intelligence.
Without a shared discovery layer, interoperability will remain bespoke, fragile, and centralized by accident.
AX is a small step toward fixing that, using the internet we already have.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article