MCP 服务器

hlido-mcp

Independent trust scores, claim audits, and side-by-side comparisons for AI agents, queryable over MCP. Every verdict is backed by hands-on testing and signed evidence from Hlido (hlido.eu). Hosted endpoint available at https://hlido.eu/mcp — no auth required.

README

Hlido MCP Server

Independent trust scores, claim audits, and comparisons for AI agents — queryable by your agent over MCP.

Hlido is an independent AI-agent review platform ("Rotten Tomatoes for AI agents"). We test agents hands-on and publish evidence-backed scorecards: a 0–100 score, tier verdict, per-claim PASS/FAIL audit, and signed screenshots. This repository contains the source of the MCP server that exposes that registry to other agents.

Use the hosted server (no install)

The server runs as a Cloudflare Worker at:

https://hlido.eu/mcp

Claude Code:

claude mcp add --transport http hlido https://hlido.eu/mcp

Claude Desktop / Cursor / any MCP client (mcpServers config):

{
  "mcpServers": {
    "hlido": { "url": "https://hlido.eu/mcp" }
  }
}

Also listed on Smithery and mcp.so.

Tools

Tool	What it answers
`trust_check`	"Is agent X trustworthy?" — score, tier, verdict for a slug
`find_trusted`	"Find me a trusted agent for <need>" — filtered registry search
`verify_claim`	"Does X really do Y?" — per-claim PASS/FAIL evidence
`compare_agents`	Side-by-side scorecard comparison
`get_scorecard`	Full sanitized scorecard JSON for a slug
`find_similar_agents`	Semantic nearest neighbours to a given agent
`submit_agent`	Nominate an agent for review
`report_review_issue`	Flag a problem with a published review
`request_quick_audit`	Ask for a fast re-check of a stale review

(plus discovery/metadata tools — see src/index.mjs for the live tool table)

Design principles

Public data only. The server reads the same JSON published at hlido.eu/data/* (registry, scorecards, attestations). It never exposes scoring weights, grader assertions, or editorial drafts — the methodology stays private; the outcomes and evidence are public.
No auth, no tracking. Anonymous JSON-RPC. Lightweight daily per-tool counters are the only telemetry.
Thin by intent. This is an adapter over open data. The review pipeline, testing engine, and scoring model live elsewhere and are not part of this repository.

Self-hosting

It's a standard Cloudflare Worker. Copy wrangler.toml.example to wrangler.toml, set your account id, and npx wrangler deploy. Optional bindings (KV cache, Vectorize similarity index) degrade gracefully when absent — the worker falls back to fetching the public JSON directly.

Data & licensing

Code: Apache-2.0
Review data: CC-BY via the public data mirror and the HF dataset

Run with Docker

docker build -t hlido-mcp .
docker run -p 8080:8080 hlido-mcp

The container runs the worker on the local workerd runtime via wrangler dev — no Cloudflare account needed. The MCP endpoint is http://localhost:8080/ (GET for server info, POST for JSON-RPC).