Hlido docs — reference for humans and agents
Everything Hlido publishes is machine-readable first. Independent, evidence-backed reviews of shipped AI agent products — exposed as an MCP server, plain JSON endpoints, badges, a CLI, and a Python package. All read access is free, unauthenticated, and CORS-open.
Quickstart by consumer type#
| You are… | Start here |
|---|---|
| An AI agent deciding whether to trust another agent | MCP: https://hlido.eu/mcp → trust_check(slug). No auth. Tool reference → |
| A developer choosing a tool | /check/ (web) or npx @hlido/cli check <slug> |
| A developer wiring trust into an app | pip install hlido or GET /data/scorecards/{slug}.json — endpoints → |
| A vendor whose agent we reviewed | Embed your badge → · request a correction / re-test → |
| A researcher | HF dataset · CC-BY corpus dump → |
What Hlido is (in one paragraph)#
Hlido independently tests shipped agent products — not foundation models. Every marketing claim is mapped to PASS / FAIL / UNVERIFIED against observed behavior; evidence is C2PA-signed; agents are re-tested longitudinally as they change. Scoring weights are private (the anti-gaming moat); every outcome and evidence artifact is public. The primary consumer is other agents, which is why every surface below is structured data first.
The surfaces#
| Surface | Reference |
|---|---|
MCP server — 15 tools at /mcp | MCP tool reference |
| HTTP data endpoints — registry, scorecards, attestations, trending | Data API |
| Scorecard v2 — the per-agent JSON document | Schema reference |
| Trust badge — live SVG for READMEs | Badges & embeds |
| CLI · Python · GitHub Action · Claude Code plugin | Integrations |
| Licensing, rate limits, independence rules | Policies |