This is a public ledger of observed contract changes across 36,574 tools in the Model Context Protocol (MCP) registry. mcpindex crawls that registry daily, detects when tool interfaces change between snapshots—things like new required parameters, changed output schemas, or flipped read/write flags—and publishes them as anonymous fingerprints with timestamps and change types. The system's core claim is that traditional auth (API keys, allowlists) doesn't catch the failure mode that actually matters in MCP: a trusted server you pinned on day 1 that silently broke its contract on day 30. The ledger acts as a public source of truth so that a "gate" (their term for a runtime guard) can warn an agent on its first call to a drifted tool. The numbers: 12,391 tools changed publishable contract fields; 7,239 in safety-relevant ways (output schema changes dominate; tool removals, type changes, and parameter shifts matter most). About 5,500 of these changes are harmless optional parameter additions. The ledger explicitly doesn't prevent calls—it just observes and warns. Absence from the ledger isn't a clean bill of health (private tools don't appear). No payment changes inclusion or exclusion. The source doesn't say what "annotation-flip-to-destructive" means, or whether agents actually react to the warnings, or what "gate" infrastructure looks like in practice.