What it checks Four groups, four stores
Each check passes (full weight), warns (half) or fails (none). The score is the weighted share over the checks that apply. Grades: A from 90, B from 80, C from 65, D from 50, F below. A store is ready when none of its blocking checks fail and every blocking check could run. If a blocking check could not run (for example, tools are behind sign-in), the store shows Unverified.
TransportHTTPS, initialize over Streamable HTTP, a supported protocol version, the response Content-Type, JSON-RPC errors, GET handling and latency.
AuthWhen the server asks for sign-in: the 401 challenge, protected resource metadata (RFC 9728), authorization server metadata, PKCE S256, client registration and refresh tokens.
Toolstools/list returns tools with object input schemas, portable names, titles, descriptions and the readOnlyHint, destructiveHint and openWorldHint annotations.
MetadataserverInfo name and version, server instructions, and the tools capability in the initialize answer.
Claude Connectors DirectoryBlocked by: no HTTPS or initialize, missing OAuth discovery or PKCE when sign-in is used, tools without object schemas, annotations or titles.
ChatGPT app directoryBlocked by the same as Claude, plus tools without an explicit openWorldHint.
Gemini Connected AppsBlocked by: no HTTPS or initialize, an unsupported protocol version, missing authorization server metadata or PKCE, tools without object schemas.
Official MCP RegistryBlocked by: no HTTPS, or no answer to initialize at the public URL.
The check connects the way an MCP client does and reads what the server publishes. See how public servers score in the McpStamp Readiness Index.