Your product might be invisible to the agents you’re building for.

Most AI-readiness tools stop at meta tags. If you’re building a product meant to be called by AI agents — not just read by them — that’s not the layer that matters. 3webs is the only audit in its category that includes the machine web: agent cards, declared capabilities, and whether they actually respond when invoked.

What the machine-web dimension checks

Agent card discoverability and validity. Is /.well-known/agent-card.json published, and does it parse as a structurally valid A2A agent card?
Capability contracts. Does every declared capability specify a real input, output and endpoint — or is it a name with nothing behind it?
Real invocation, not just declaration. For any capability marked safe to test, the engine actually calls it over JSON-RPC 2.0 and checks the response — the same distinction between "documented" and "working" that bites every API integration.
Task and state contracts. Human-approval boundaries, task status, cancellation and resume — the parts of the A2A spec that are easy to leave half-implemented under deadline pressure.

Why this matters now

Agent-to-agent protocols are early enough that most published agent cards have never been tested against a real client. A capability that looks complete in capabilities.json but 500s on invocation is invisible the moment an agent actually tries to use it — and unlike a broken webpage, nobody files a bug report.

Full transparency

3webs runs this exact audit against itself, live, at 3webobs.com/self-audit.html — the same engine, the same rules, no special-casing for our own domain. If you want to see what a "capability invocable" pass actually requires, that page shows it in real time.

Check whether your own agent card and endpoints actually work.

Run a free audit →