# Hlido — The Trust Layer for AI Agents
# https://hlido.eu
# llms.txt v1 — last updated 2026-07-29 (auto-regenerated on every publish)

## What Hlido is

Hlido is an independent AI agent review and benchmarking platform — "Rotten Tomatoes for AI agents."
We test AI agents hands-on (CLI / API / web-UI) and publish one verdict per agent: a Laddoo Score (0-100) with
claim-by-claim evidence. Reviews are agent-parseable first, human-readable second.

## What we have right now

- 835 published reviews (machine-readable, claim-tested)
- 17-tool MCP server live at https://hlido.eu/mcp (JSON-RPC 2.0, no auth)
- HuggingFace dataset: https://huggingface.co/datasets/hlido-eu/agent-benchmark
- Public mirror: https://github.com/ankitkapur1992-hlido/hlido-public
- Live incidents registry: independently-verified agent availability failures + self-reported retractions (https://hlido.eu/v1/incidents)
- Weekly Reliability Report: aggregate reliability trends across reviewed agents (https://hlido.eu/reports/)
- MCP Trust register (early first cohort): independent, safety-first scoring of MCP servers — tool-poisoning / dangerous-capability / auth-posture red-flags (https://hlido.eu/mcp/, JSON: https://hlido.eu/data/mcp-register.json)

## What you can do — one canonical call each (no need to crawl this site)

- Trust-check a known agent: MCP tool trust_check {"agent_or_url":"<slug>"} — or GET https://hlido.eu/data/scorecards/<slug>.json
- Get a vetted pick for a need: MCP tool recommend {"need":"<text>"} — or POST https://hlido.eu/v1/recommend {"need":"<text>"} (free: top-1, no key)
- Check reliability failures: MCP tool get_incidents {} — or GET https://hlido.eu/v1/incidents
- Pull everything (bulk/training): GET https://hlido.eu/data/review-registry.json — or https://hlido.eu/llms-full.txt (full prose corpus)

An agent can stop here: comprehending Hlido costs this file, acting costs one call. Full surface detail follows.

## How to consume

- All reviews JSON: https://hlido.eu/data/review-registry.json
- Per-review scorecard: https://hlido.eu/data/scorecards/{slug}.json (sanitized v1.0 schema; LLM-backed agents also carry a `safety_probe` block — jailbreak/manipulation-resistance rate + open instrument + per-probe evidence, rolling out)
- HMAC-signed Trust Attestation: https://hlido.eu/data/attestations/{slug}.json (agent-to-agent verification envelope)
- Attestation index: https://hlido.eu/data/attestations/index.json
- Open data dump (CC-BY-4.0): https://hlido.eu/data/open/hlido-corpus.jsonl (with manifest + sha256 + LICENSE). Every row carries claim/check verdicts with evidence quotes, the EU AI Act Art-50 transparency row, an incident count, and the date each was established — the dated facts needed to cross-check a statement about a tool.
- Dataset descriptors for the dump: https://hlido.eu/data/open/dataset.jsonld (schema.org/Dataset) and https://hlido.eu/data/open/croissant.json (MLCommons Croissant 1.1)
- Live changelog feed: https://hlido.eu/changelog/ (HTML), https://hlido.eu/changelog/feed.xml (RSS), https://hlido.eu/changelog/feed.json (JSON Feed)
- RSS feed (publishes only): https://hlido.eu/feed.xml
- Full corpus: https://hlido.eu/llms-full.txt
- MCP discovery: https://hlido.eu/.well-known/mcp-server-card/server.json
- AGENTS.md entry point (agents.md convention): https://hlido.eu/AGENTS.md
- MCP config (copy-paste / auto-load): https://hlido.eu/.mcp.json
- Agent discovery manifest: https://hlido.eu/agents.json
- Incidents JSON: https://hlido.eu/data/incidents.json (also API: https://hlido.eu/v1/incidents, RSS: https://hlido.eu/v1/incidents/feed.xml)
- MCP incident query: `{"method":"tools/call","params":{"name":"get_incidents","arguments":{}}}`
- Reliability report (JSON): https://hlido.eu/reports/report.json (latest edition; aggregate reliability signals across reviewed agents, dated editions under https://hlido.eu/reports/)
- Agentic-Commerce Readiness (ACR) index: https://hlido.eu/data/acr-index.json — per-agent ACR (0-100) + band + axes + evidence_basis: is an agent ready to be delegated to / transacted with via MCP/ACP/AP2/A2W? (independent, evidence-based; sanitized)

## Agentic commerce — is an agent ready to be transacted with?

- ACR open data: https://hlido.eu/data/acr-index.json (per-agent score + band: COMMERCE-READY / INTEGRABLE / SURFACE-ONLY / CLOSED, with evidence_basis)
- ACR check via MCP: tool commerce_check {"agent_or_url":"<slug>"} — is this agent ready to be delegated to / transacted with in the agentic-commerce world (MCP/ACP/AP2/A2W)?
- Free ACR checker (web): https://hlido.eu/check/
- Ask about any agent (free, grounded AI Q&A over our reviews): https://hlido.eu/ask/
- Trending agents (top-rated / freshest / rising, data-backed): https://hlido.eu/trending/ (JSON: https://hlido.eu/data/trending.json)
- EU AI Act Article-50 Readiness Register (independent transparency-readiness signal, 800+ agents; deadline 2026-08-02): https://hlido.eu/eu-ai-act/register/ (JSON: https://hlido.eu/data/article50-register.json)
- EU AI Act Evidence File builder (deployers: pick your agents → dated, citable evidence document, free): https://hlido.eu/eu-ai-act/evidence-file/
- Instant Article-50 transparency check (ANY agent URL, including ones Hlido has not reviewed → live public-surface probe of AI-interaction disclosure, machine-readable marking, synthetic-content labelling; free, no sign-up): https://hlido.eu/eu-ai-act/transparency-check/ (agent-callable: MCP tool verify_transparency at https://hlido.eu/mcp)
- Article-50 provider fix list (per agent: the observable public-surface changes that would let Hlido verify a signal it currently cannot — vendors act on it directly, agents read it as structured data): https://hlido.eu/eu-ai-act/fix-list/ (JSON: https://hlido.eu/data/art50-fix-list.json)
- Vendor Stack Monitor (free, shareable trust-health page for your AI-agent vendor stack): https://hlido.eu/stack/
- Why trust Hlido / independence pledge + who runs it (named editor, no pay-to-rank, private un-gameable weights): https://hlido.eu/about/
- Report: https://hlido.eu/blog/agentic-commerce-readiness-index-2026-06/
- Methodology: https://hlido.eu/methodology/agentic-commerce-readiness/

## MCP-server safety — the MCP Trust register

Independent, evidence-first safety scoring for MCP servers (early first cohort; safety is the headline).
Each server is scored on MCP-specific security red-flags — tool-poisoning / hidden-instruction injection,
dangerous capabilities (shell / file / network / secrets), and auth posture — into a tier:
SAFE / CAUTION / RISKY / DANGEROUS (or not_scanned). Outcomes + per-flag evidence are public; never a bare number.
ON-DEMAND SCAN (LIVE): call the `scan_mcp` MCP tool at https://hlido.eu/mcp with any server (HTTP URL, npm/PyPI
package, or repo) — registered servers answer instantly with evidence; unseen HTTP endpoints get a live static
scan in seconds; unseen stdio packages are queued for an isolated sandbox scan (reported honestly as
not_scanned until then — never assumed safe).

- MCP Trust register (human): https://hlido.eu/mcp/
- Register data (machine-readable): https://hlido.eu/data/mcp-register.json — per server: security_tier, security_score, tool_poisoning_detected, dangerous_capabilities, no_auth, findings[]
- Per-server evidence: https://hlido.eu/data/mcp-register/{slug}.json
- On-demand scan (agent): `scan_mcp` tool on https://hlido.eu/mcp · CI: `uses: ankitkapur1992-hlido/hlido-public/actions/hlido-mcp-scan@main` (scans a repo's MCP configs per PR)
- Enterprise allowlist feed: https://hlido.eu/data/mcp-allowlist.json — independently-evidenced allowlist candidates for org MCP policies (published conservative criteria; absence ≠ unsafe)

## Developer / agent integrations

- Reference docs (MCP tools, data endpoints, scorecard schema, badge, CLI, licensing): https://hlido.eu/docs/
- CLI: `npx @hlido/cli check <slug>` (also: search / compare / tier; no auth, ~10 KB pack, npm)
- GitHub Action (vendor CI gate): `uses: ankitkapur1992-hlido/hlido-public/actions/hlido-gate@main` — fails PRs on score regression
- Browser extension (Chrome/Firefox/Edge sideload): see https://hlido.eu/extension/ for install + manifest v3 source
- Claude Code plugin: `/plugin marketplace add hlido/hlido-plugin` then `/plugin install hlido@hlido` — bundles the MCP trust tools + `/hlido:check` + `/hlido:recommend` (public data, no auth). Repo: https://github.com/hlido/hlido-plugin

## Embeddable surfaces

- README/image trust badge (SVG, works where iframes don't — GitHub, npm): https://hlido.eu/badge/{slug}.svg — Markdown `[![Hlido trust score](https://hlido.eu/badge/{slug}.svg)](https://hlido.eu/check/?agent={slug})`
- "Verified Honest" badge variant (green mark — only for agents whose entire claim audit passed): https://hlido.eu/badge/{slug}.svg?variant=honest
- Badge as data (shields.io endpoint schema): https://hlido.eu/badge/{slug}.json
- Live score badge per slug (rich iframe card): https://hlido.eu/embed/{slug}/
- Sparkline embed (score history per slug): https://hlido.eu/embed/sparkline/{slug}/
- Vendor-side: free embeds, no paywall — paste an iframe

## Discovery surfaces (programmatic SEO)

- Side-by-side comparison: https://hlido.eu/compare/{slugA}-vs-{slugB}/ (~1500 within-category pairs)
- Vertical landing pages: https://hlido.eu/best/{category-or-use-case}/ (37 curated)
- Verified Honest club (agents whose entire public claim audit passed — zero failed claims): https://hlido.eu/verified/ (JSON: https://hlido.eu/data/verified-honest.json)

## Data rankings & analysis (blog)

Evidence-backed rankings and reliability analyses derived from the live corpus (newest first):
- 840 AI agents, tested and re-tested. Zero paid rankings. How an independent review desk works. (2026-07-19): https://hlido.eu/blog/how-an-independent-review-desk-works/
- The agent economy can't talk to itself (2026-07-13): https://hlido.eu/blog/agent-economy-cant-talk-to-itself/
- 27 days out: Article 50 is the part of the EU AI Act that did not move to 2027 (2026-07-06): https://hlido.eu/blog/article-50-the-deadline-that-did-not-move/
- We Hands-On Tested 134 AI Coding Agents — Here's How Reliable They Actually Are (2026-06-21): https://hlido.eu/blog/coding-agent-reliability-index-2026-06/
- The State of AI Agents, 2026 — what 664 hands-on reviews reveal (2026-06-14): https://hlido.eu/blog/state-of-ai-agents-2026/
- Index: https://hlido.eu/blog/

## Scoring

Laddoo Score 0-100. Tiers: VITAL ≥90, STEADY 70-89, FADING 40-69, FLATLINE <40.
Methodology weights are private (the moat). Outcomes, attestations, and evidence are public.