I built this because robots.txt kept coming up in conversations about AI crawlers
and the answer was always "it's a polite request". AIFeed tries to make it checkable.
What it is: a publisher signs a small JSON manifest with Ed25519 and anchors the
key in DNS. Agents verify TLS -> domain -> signature -> DNS, then re-check a revocation list. Two markdown profiles for delivery, plus a signed delta index so
unchanged pages cost nothing.
Publishing a static site is one command: npx aifeed init --domain example.com --dir ./site Verifying any domain is three lines with @aifeed/verify.
Numbers from the local harness: -68.8% bytes vs HTML (-95.7% with the delta index), 0.70 ms per signature check. It's a simulation harness, not a field trial; the
30-day pilot hasn't run.
Honest limits: no external crypto review yet, the specs are a draft, and
origin+DNS compromise on first contact is undetectable.
Happy to answer questions about the design (JCS+Ed25519, DNS anchoring,
revocation semantics, MAKO compatibility).
[https://aifeed.md](https://aifeed.md)
Run: npx -y aifeed-mcp-server
Registry: io.github.denyn1/aifeed-mcp-server
Smithery: [https://smithery.ai/server/denyn1/aifeed-mcp-server](https://smithery.ai/server/denyn1/aifeed-mcp-server)
Repo: [https://github.com/denyn1/aifeed-protocol](https://github.com/denyn1/aifeed-protocol)