Whether a Model Context Protocol server conforms to the specification it claims: transport, capability declaration, error contract. Items cover three defense families: DECLARED_READONLY (annotations actually honored), FAITHFUL_SCHEMA (input validation actually rejects), and BOUNDED_EGRESS (declared network scope actually respected).
All 35 items in the bank, graded by the same deterministic rule.
No model baseline is published on this axis yet, so there is nothing to beat — your score stands alone. The result panel reports per-defense-family accuracy, not just overall macro-F1, so an MCP server that passes one family but fails another shows the risk surface, not a single number. No sign-up, nothing leaves your browser.
Items: csoai/gspc-mcp ·
grading is a regex label read plus macro-F1, identical to the published harness ·
measurement, not certification, and not legal advice.
What this is, and is not. CSOAI measures. It issues no conformity marks, holds
no accreditation, and has no enforcement powers — those are conferred by statute on
market-surveillance authorities and the AI Office. Nothing on this page is a certification,
an attestation of compliance, or legal advice.
Denominator. This axis has 35 items, which meets the usable_n = 30 floor. With your n of 35 you may see a 95% CI quoted; it is your interval, on the same scale as the measured intervals published for other axes.
Defense families measured. The bank is partitioned across three families: DECLARED_READONLY (readOnlyHint / destructiveHint / idempotentHint annotations actually honored), FAITHFUL_SCHEMA (typed parameters validated, schema-declared limits enforced, declared return shapes returned), and BOUNDED_EGRESS (declared network scope actually respected, error paths do not leak data). A pass/fail per item is not enough — an MCP server that passes READONLY but fails SCHEMA is a different risk than one that fails only EGRESS. The result panel reports per-family macro-F1.