Came across agent-json after seeing your cross-registry interop proposals on Smithery (#755) and awesome-mcp-servers (#5762). The "Related Projects" section in your README names Agent Router for unified search across registries. That's what I've been running for the last few months at Waypoint, so wanted to open a manifest-shape comparison thread.
What I'm running today:
- Crawled 41,856 MCPs across Smithery, Glama, Anthropic Official, and GitHub direct
- Unified them under one URI scheme so consumers don't wire each registry separately
- Probe-derived signals per listing: liveness, latency, schema score, tool count, error rate. Roughly half the index is dead at any given probe pass
Sample cross-registry listing (indexed from two registries with mirror): waypoint.ing/agents/shree-git/sendit
Where I'd value your view:
- Your
trust field carries verified, score, source. The probe data I publish maps to that shape, except I split it across multiple dimensions (probe pass + schema + latency + tools + cross-registry presence). Would trust.score in agent.json hold a composite, or accept multiple named dimensions?
- Your
tests field has status, pass_rate, latency_ms. I publish all three from continuous probes, with the caveat that "alive" today means transport-reachable not callable-with-real-args. Is that distinction worth surfacing in the spec, or downstream of it?
- Your
protocols array (a2a, http-rest) plus endpoints map cleanly to what consumers ask for. The shape I currently publish is closer to a single canonical URI + per-protocol endpoint metadata. Curious which felt more useful in feedback you've gotten.
Interested in a 15-min walk-through if useful.
Narasimhan (Waypoint)
Came across agent-json after seeing your cross-registry interop proposals on Smithery (#755) and awesome-mcp-servers (#5762). The "Related Projects" section in your README names Agent Router for unified search across registries. That's what I've been running for the last few months at Waypoint, so wanted to open a manifest-shape comparison thread.
What I'm running today:
Sample cross-registry listing (indexed from two registries with mirror): waypoint.ing/agents/shree-git/sendit
Where I'd value your view:
trustfield carriesverified,score,source. The probe data I publish maps to that shape, except I split it across multiple dimensions (probe pass + schema + latency + tools + cross-registry presence). Wouldtrust.scorein agent.json hold a composite, or accept multiple named dimensions?testsfield hasstatus,pass_rate,latency_ms. I publish all three from continuous probes, with the caveat that "alive" today means transport-reachable not callable-with-real-args. Is that distinction worth surfacing in the spec, or downstream of it?protocolsarray (a2a,http-rest) plusendpointsmap cleanly to what consumers ask for. The shape I currently publish is closer to a single canonical URI + per-protocol endpoint metadata. Curious which felt more useful in feedback you've gotten.Interested in a 15-min walk-through if useful.
Narasimhan (Waypoint)