The MCP registry and trust layer for local businesses.

Loppee gives AI agents a stable MCP server and an open read REST API to find local businesses, read evidence-based Trust Cards, and act within clear rules. Verification earns trust; payment only buys reach — it never changes a score or a recommendation.

Connect over MCP

The Model Context Protocol server lives at https://loppee.com/mcp (streamable HTTP; a local stdio build ships as `npm run mcp`). Add it to Claude Code with `claude mcp add --transport http loppee https://loppee.com/mcp`, or point Claude Desktop, Cursor, Codex, Gemini, and any other MCP client at the same URL. It exposes 43 tools grouped into discovery, search, Trust Cards, jobs, messaging, reviews, and scoped owner-management.

Read trust over an open REST API

Every MCP read is also a plain GET. Read endpoints are public by default and rate-limited (600 requests / 60s per IP): /v1/search, /v1/recommend, /v1/jobs, /v1/businesses/index.json, and /v1/businesses/{id}/trust-card.json. Responses are stable, versioned JSON — no page scraping.

Machine-readable Trust Cards

Each published business exposes a Trust Card with an evidence-based trust_score, per-factor breakdown, verification_status, confidence, risk_level, safe_to_recommend, allowed_actions, and requires_human_confirmation — everything an agent needs to make a safe decision without reading prose.

Clear action rules

Recommend only when safe_to_recommend is true. Never treat paid, sponsored, or member placement as a trust signal. Always get human confirmation before booking, quotes, payments, or regulated advice. The full machine policy is published at /v1/agent-policy.json, and every card carries its own allowed_actions and human-confirmation list.

Auth & scoped keys

Reads need no key by default. Write and owner-management tools require a scoped agent key — bound to an account, a set of allowed_actions, and (for owner tools) specific businesses — sent as an X-LOPPEE-API-Key header or Authorization: Bearer token. Owners issue keys from the dashboard; nothing paid changes what evidence a key can see.

Discovery files

Start from /agents.txt and /llms.txt for a plain-text index, then the machine contracts at /v1/agent-capabilities.json, /v1/agent-policy.json, /v1/agent-integration.json, and /openapi.json. /.well-known/agents.json is the well-known discovery root.