| MCP Registry presence | Whether your server is listed in the official MCP Registry with a resolvable entry — the front door AI clients use to discover servers. | Scored |
| Directory & listing coverage | Presence across the syndicated directories agents and humans browse — Glama, PulseMCP, Smithery — plus your package-registry entry and source repo. | Scored |
| Tool metadata quality | A live MCP handshake (initialize + tools/list) reads your real tool names, descriptions, and input schemas, then scores how clearly they describe what each tool does — the text a model reads when deciding whether to call you. | Scored |
| Directory-review compliance | Alignment with the published Anthropic (connectors) and OpenAI (apps) directory-review rubrics — the criteria a human reviewer applies before listing you. | Scored |
| Tool-search retrieval | A deterministic BM25 simulation of the lexical tool-search stage AI clients run before model selection: given realistic queries, does your tool surface? The same target always produces the same result. | Scored |
| OAuth / security / trust surfaces | Presence and shape of trust signals: OAuth protected-resource metadata, /.well-known/agent-card.json, security.txt, and Web Bot Auth key directories. We report what exists; we are not a security scanner. | Scored |
| Agent badge & verify page | A public, payment-independent verify page and an embeddable live-grade badge for your server — a corroborating, third-party-readable trust surface you can link from your repo and listing. | Scored |
| Forward-looking standards | Emerging agent standards — A2A agent cards, Web Bot Auth, and agent-payment signals (x402 / AP2). Shown for roadmap value and labelled advisory; never counted against your grade. | Advisory |