Sealmarkdocs methodology 0.3-draft
methodology 0.3-draft

Documentation

Sealmark runs fixed probes against an MCP server and grades what it does, not what it says.

shipped

Sealmark starts your MCP server, calls its tools directly with fixed inputs, and records what happened. No LLM is in the loop. Where it can, it checks the database itself instead of trusting the server's reply. It is behavioral conformance testing for MCP servers you run or build, not a security scanner and not a certification.

What you get#

  • One letter per pack, with the evidence behind every check line. A grade is a local test result for one server version in one configuration: read the per-check lines, not just the letter.
  • A JSON record of the run, so a result can be reproduced and diffed, plus other report formats.
  • An exit code for CI: any graded pack below an A exits 1 by default; see Run in CI.

Start with the Quickstart. Anything labelled planned is graded by nothing.