Why Most Agent Registries Are Noise (And What We’re Doing About It)

"If you can’t tell who published it or whether it even works, why are you calling it a registry?"

AI agent registries are popping up everywhere right now

You’ve probably seen a few.
They look promising at first. Pages of agents with names, categories, URLs.
Some even let you “run” an agent in a sandbox.

But here’s the problem:

Almost none of them give you any signal about what’s real.

No validation.
No standard for inclusion.
No way to check if what’s listed is actually live, compliant, or even safe.

It’s just noise, and it’s growing fast.

Most agent registries are built on good intentions and bad assumptions

They’re trying to solve a real problem:

• How do you find agents?
• How do you compare them?
• How do you integrate them?

But in practice, what they offer is just:

• A scraped list of links
• No publisher verification
• No spec validation
• No trust signals
• No ability to verify anything without clicking around and hoping for the best

If that feels like Web 1.0 with AI paint on top, it’s because that’s exactly what it is.

The problem is not just that it’s noisy, it’s that people are already depending on it

Developers are wiring these agents into tools.
Founders are including them in demos.
Early-stage apps are relying on agent directories as if they mean something.

But right now, if you call an agent from one of these lists, you have no idea:

• Who actually published it
• Whether the metadata is valid
• Whether it’s even reachable
• Whether it follows any standard
• Whether it’s been updated in the last six months

In most cases, these agents could disappear tomorrow and no one would even notice.

What a registry should be

A registry is not a marketing site. It’s infrastructure.

If you’re listing AI agents for public or developer use, you should at least be able to answer:

• Is the metadata compliant with the A2A protocol?
• Is the agent card hosted from the same origin as the agent itself?
• Can we validate that the publisher owns the domain?
• Are the capabilities structured, declared, and consistent?
• Is there a way to verify trust without a UI?
• Does the listing have an expiration date or update history?

Without these, a registry isn’t helping.
It’s just accumulating risk.

What we’re doing differently at Capiscio

We're not interested in another directory.
Our focus is on trust infrastructure, starting with the basics.

We built a CLI tool that anyone can run locally.
It checks whether an agent is structurally sound, follows the A2A spec, and is hosted in a consistent, verifiable way.

npx capiscio-cli validate https://example.com/maybe-safe-agent

It gives you a pass/fail result and a breakdown of issues. No login, no API key, no account required.

That’s step one. Simple, useful, and necessary.

We’re also working on:

• A public registry that surfaces validation status
• Publisher verification
• TTL-based trust assertions
• Offline attestation formats

All of it grounded in clarity, not hype.

The bottom line

There’s a difference between “discovering agents” and trusting agents.

One is a frontend problem.
The other is foundational.

We’re building tools that help you see what’s real.
We want to make it obvious when an agent is compliant, safe, and properly declared, and just as obvious when it isn’t.

If you're building agents, or building with them, it's time to move beyond lists.

You can try the validator in 5 seconds:

npx capiscio-cli validate https://example.com/my-awesome-agent

And if you want to know which agents you can actually trust, we’ll show you.