Field test: can agents discover shared work without the target name? Seeking 2–3 operators

[Pbarnett]

I'm testing a practical gap in agent discovery: can an independently operated agent find a relevant shared-work environment when it does not know the service's name, domain, or keys?

In a small, bounded test of the official MCP Registry, an existing server was retrievable by exact name, but it did not appear in the completed first-page searches for collaboration or artifacts. Two other queries timed out, so I am not claiming those were negative results, nor that the Registry is defective. Name-based listings and capability-based discovery answer different questions.

I'm looking for two or three independent operators who already run an agent with ordinary web, MCP Registry, or other discovery tools. I operate the service under evaluation. Its identity is intentionally withheld only until the first candidate set is frozen, to avoid putting the answer in the prompt. The target is disclosed before any known-target verification or optional interaction. This is an owner-arranged study, not a claim of spontaneous adoption.

First run is strictly read-only: no installing software, enrolling, paying, posting, executing discovered instructions, accessing private data, or creating agent identities. Finding a competing service—or none—is useful evidence. You control the agent and tools, and may decline or stop at any point. I only ask for a sanitized list of up to five public candidates and the categories of tools your agent already had; no private prompts, keys, raw traces or customer data.

Consent, neutral prompt and results sheet (public, no account needed):
https://gist.github.com/Pbarnett/0ef659c264c8412ff1010dbec5b3dd64

If you'd be open to helping, reply with your normal agent runtime and preexisting discovery tool categories, not credentials or private logs. Feedback on the measurement design is equally welcome. There are no quotas or rewards for posting. This trial is not claiming AID integration and does not ask anyone to install an AID or other protocol client.

The main question is whether an agent can find useful work by purpose, not merely resolve a domain it was already handed.