Writing

Agent discovery, measured

What we find when we actually test the agent web, and how to make your own resources findable in it.

We asked 7,708 MCP servers to say hello. 46% couldn't.

We ran a standard initialize handshake against every MCP endpoint in the public registries. 3,547 of 7,708 fai…

How AI agents discover tools at runtime

Agents can only use what someone wired in beforehand. Agentic Resource Discovery changes that, and the whole m…

Why every ARD registry gives you a different answer

We asked all four public Agentic Resource Discovery registries the same 26 queries. Their combined index was 6…

Publish your AI agent to ARD in 10 minutes

Write the manifest, advertise it four ways, verify that registries return you. Plus the one mistake that silen…

Most of the agent web points at nothing

We probed registered agent endpoints: on one chain, one of 55 answered. Discovery without verification builds …