Hlido is the trust layer for AI agents. When an agent is about to call another agent — or when a buyer is choosing between AI products — Hlido provides the pre-flight check: independent reviews backed by C2PA-signed evidence, machine-readable claim-vs-evidence scorecards, and a Laddoo Score that's consistent across every reviewed agent.
The MCP server exposes 9 tools for both human and agent consumers:
-
trust_check — single-call verdict (slug or URL → score, tier, blockers)
-
get_scorecard — full sanitized claim-vs-evidence scorecard, schema v1.0 (every command, every exit code, every source surface)
-
find_similar_agents — semantic search over the corpus via OpenAI embeddings + Cloudflare Vectorize ("I need an agent that can refactor TypeScript and edit multiple files" → top-N matches)
-
find_trusted — substring search across reviewed agents
-
verify_claim — checks whether a Hlido review references a specific claim (honest nulls when not tested)
-
compare_agents — side-by-side comparison of up to 5 agents
-
request_quick_audit — request a public-surface audit for an unreviewed agent (rate-limited: 5/day anonymous, 50/day identified). Returns a future scorecard URL + ETA
-
report_review_issue — flag a stale verdict, wrong claim, or broken link. Routes to dispute-retest queue
-
submit_agent — slow-queue submission for daily intake
No auth required. JSON-RPC 2.0. Public, free, read-only. Reviews regenerate per-publish; the index is current within 5 minutes of any update.
Why we exist: in 2030 most "is this AI agent reliable?" questions will be asked by other agents, not humans. Hlido is the cryptographically-signed trust layer they query. Every review carries a real run_id, a real terminal cast (CLI agents) or signed PNGs (web agents), and a structured scorecard that any consumer can parse without scraping HTML.
Server Config
{
"mcpServers": {
"hlido": {
"type": "streamable-http",
"url": "https://hlido.eu/mcp"
}
}
}