MCP Compatibility inspects a public MCP endpoint and records what a standards-aware client can actually observe. It keeps transport, server metadata, authorization discovery, tool schemas, annotations, latency, and incomplete evidence separate so one missing feature never becomes a misleading overall grade.
The check uses bounded requests, a forty-second whole-operation deadline inside its forty-five-second execution profile, and does not invoke any remote tool. Each result records the observation time and enforced limits, which makes it useful for release checks, interoperability debugging, and evidence-backed documentation.
What you can do
- check_server — negotiate an MCP session and inspect one bounded tools/list response
- Distinguish Streamable HTTP, legacy SSE, authorization-required, unavailable, and incomplete states
- Review raw schema and annotation counts without an opaque compatibility score
- Capture explicit timeout, redirect, response-size, and truncation evidence
Who it's for
MCP server maintainers, client developers, registry publishers, and platform teams checking interoperability. It is designed for public endpoints that can be tested without sending credentials.
How to use it
- Call check_server with the exact public MCP endpoint
- Review transport and metadata before interpreting the tool evidence
- Treat any incomplete observation as a reason to retry or investigate, not as proof of absence
- Preserve the observation timestamp and limits when sharing results
Getting started
Pass one public HTTP or HTTPS MCP endpoint and adjust the bounded limits only when the server legitimately needs them.
