MCP Declared-Effect Coverage and Contract Binding v1

Declared-effect coverage and contract binding for the public Model Context Protocol registry: 35 crawl observations, 2026-06-09 to 2026-08-01, covering 44,172 tools on 2,043 reachable servers. Per tool: binding state (bound / stale / unobserved / undeclared), declared MCP effect-hint vector (readOnlyHint, destructiveHint, idempotentHint, openWorldHint), observation spans, and 1,186 corroborated declaration-flip records with per-hint direction (permissive / restrictive / first-declaration / retraction), including 397 destructiveHint flips. Headline findings: 83.8% of tools declare at least one effect hint; 59.3% hold a declaration still bound to an unmutated contract (binding gap 24.5 percentage points); 783 confirmed flips of the declared class itself. Server and tool identifiers are pseudonymized with change-preserving HMAC-SHA256 tokens sharing the token space of the MCP Registry Drift Panel v1 (10.5281/zenodo.21751273), so records join across the two datasets. Every number regenerates from the released files; see DATASET_CARD.md and REPORT.md for definitions, corroboration semantics, scope limits, and the pre-publication correction note.