MCP-Server Trust Register
Before you wire an MCP server into your agent, know what it can do if it's hijacked. We independently check each server for tool-poisoning, dangerous capabilities (shell, file-write, network, secrets), and auth posture — and we show the red-flags, not just a number. No vendor pays for placement.
SAFE
CAUTION
RISKY
DANGEROUS
NOT SCANNED
Loading the register…
Agents & tools: machine-readable at GET https://hlido.eu/data/mcp-register.json
Server not listed? Scan it on demand — live. Agents: call the
scan_mcp tool on https://hlido.eu/mcp with any server URL or package name. Developers: npx @hlido/cli scan <server>, or gate your repo's MCP configs in CI with the hlido-mcp-scan GitHub Action. Registered servers answer instantly; unseen HTTP endpoints get a live static scan in seconds; local (stdio) packages are queued for an isolated sandbox scan — reported honestly as NOT SCANNED until the verdict lands here (typically <24h). Free, no auth.
How to read this. The tier reflects blast radius if the server is hijacked, not whether the maintainer is trustworthy — a file-writing or shell-executing server is powerful and risky by nature, so
RISKY can be the correct rating for a legitimate, popular server. Tool-poisoning (hidden instructions in a tool's description) is the one flag that signals possible malice. NOT SCANNED means we couldn't start the server to test it — never assume it's safe. This is an early, growing cohort; ratings refresh as servers change. Methodology is public and the same for every server.