A remote MCP server's tool names, descriptions and input schemas are text a model reads as trusted instruction. This page reads what the public transparency log recorded for an endpoint.
HISTOR reads the tool list of every remote MCP server in the official registry, pins a digest, and publishes the record. The text is then run through WARDEN, a static scanner with two tiers:
block — a pattern with no plausible benign reading in a tool definition. advise — a pattern that legitimately appears in honest tools (a password hashing tool has to receive a password). Advisory findings never block.
A block is not a verdict of malice, and it is not always right. Two errors are common: domain vocabulary that reads as a secret (mnemonic is also an EVM opcode name), and a genuinely bad design (a tool that asks the caller for a wallet private key). We wrote about both, with the registry-wide counts, in 1,085 of 103,271 MCP servers carry a block-tier finding.
histor.modelmarket.dev in your browser. We do not proxy it and we do not log the endpoint you type — the log operator does see the request, and its own answer says what it counts. No part of this page is a claim that a server is safe or unsafe; it is a reading of someone else's public record.