Re: Discussion on agent discovery mechanisms — do you have any better proposals?

Hi,
I like the idea that agents exchange information and build up knowledge of their network, however I feel agent discovery starts with finding what to connect with, a bit like linked in.

In discussions I have had, trust seems to be central to the question of agent discovery mechanisms before considering what is discovered. Centralised registries certainly provide mechanisms to find information and they were missing from the early internet, a void quickly filled by directories (AltaVista) and search engines(Google, EuroFerret, etc). The weakness of a centralised registry that holds complete metadata records (eg whole cards) is it centralises responsibility, and therefore trust. We have seen recently from registries like npm (on which IIUC mcp registry is loosely proposed) that untrusted content from malicious actors damages the authority of the registry and needs constant work at scale to maintain, especially if the registry requires owners push. So information held centrally (approach 2) is always dependent on that authority. The greater the cardinality, the harder it becomes.

I think the mcp registry model [1] is addressing this with pointers to authoritative sub registries, which perhaps if adopted here could consent pointers to .well-known locations (approach 1). Each pointers authority being established by both DNS ownership and the quality of the TLS certificate, building on the same delegation that underpins the internet today.  Registry information then becomes only as authoritative as the registry it is held in, with the ultimate source of authority being the .well-known location on the server hosting that path.

Ie, Proposal:
use approach 2 to hold informative pointers to approach 1 to hold authoritative information.
If organisations feel there is value in providing a service 2, then they can without the drawbacks of being the sole source of truth. I have a feeling these already exist. (AltaVista for Agents)




Best Regards
Ian


1 https://github.com/modelcontextprotocol/registry

--

Privacy information is available at:
https://www.adobe.com/content/dam/cc/en/legal/documents/Supplement-communications-global-workforce.pdf

From: Adam Sobieski <adamsobieski@hotmail.com>
Date: Wednesday, 8 October 2025 at 05:13
To: Gaowei Chang <chgaowei@gmail.com>, public-agentprotocol@w3.org <public-agentprotocol@w3.org>
Subject: Re: Discussion on agent discovery mechanisms — do you have any better proposals?


EXTERNAL: Use caution when clicking on links or opening attachments.


Gaowei Chang,
All,

Hello. I would like to share some ideas to add to the listed agent discovery approaches. With respect to end-users' AI agents, e.g., their personal assistants, one can envision using something resembling vCards (as attachments on email or other messages) for introducing their AI agents to one another.

In this approach, end-users (or their AI agents) would provide attachments on email or other messages. These attachments would be intended for use by recipients' AI agents. This would enable a kind of agent introduction or discovery.

Information about end-users' contacts' AI agents could be stored in their "digital rolodexes", perhaps in their People or Contacts apps. As considered, contacts could have multiple AI agents.

New capabilities and features can be envisioned, benefitting end-users, co-workers, and teammates, when their AI agents, e.g., their personal assistants, could more readily contact and communicate with one another.

Thank you. Hopefully this message-attachment-based agent discovery approach is of some interest to the group.


Best regards,
Adam Sobieski

________________________________
From: Gaowei Chang <chgaowei@gmail.com>
Sent: Thursday, September 25, 2025 10:10 PM
To: public-agentprotocol@w3.org <public-agentprotocol@w3.org>
Subject: Discussion on agent discovery mechanisms — do you have any better proposals?


Dear all,

I originally wanted to discuss the issue of agent discovery at our last meeting, but we ran out of time. Let’s continue the discussion here by email.. I have outlined three main approaches and would like to hear your thoughts:

1. Based on RFC 8615 (.well-known path)

Place a standardized file under the domain’s /.well-known/ path to declare the agents available under that domain.

  *   Pros: Mature standard, easy to deploy, compatible with DNS/TLS, decentralized.

  *   Cons: Limited to existing domains, lacks global indexing, less friendly for individual users without domains.

2. Global Registration Center

Establish a centralized registry for agents, such as an MCP Registry or an Agent Name Service (ANS).

  *   Pros: Strong discoverability, good user experience, standardized naming and classification, easier governance.

  *   Cons: Higher centralization risks, requires governance and maintenance, may introduce entry barriers, scalability challenges.

3. Blockchain-like Decentralized Approach

Use decentralized infrastructures such as blockchain, DHT, IPFS, or ENS to store and discover agent information.

  *   Pros: Decentralized, censorship-resistant, data integrity, global discoverability, can integrate with DID/VC systems.

  *   Cons: Complex to implement, performance and cost issues, ecosystem still immature.

Which approach do you prefer?


Best regards,
Gaowei Chang