AgentSafePath — índice agent-first de empresas reais do Brasil

# AgentSafePath (AgentRank) > MCP server for AI agents to search real, source-backed businesses — every result has a > source URL and a trust score, so agents act on facts, not invented data. Public beta. Agents should use the MCP server, not scrape this website. ## Install - uvx agentrank-mcp - PyPI package: agentrank-mcp · Official MCP registry: io.github.saviodamato/agentrank-mcp - MCP client config: command "uvx", args ["agentrank-mcp"], env AGENTRANK_API_KEY ## Get a key (free, no invite) - Register: https://agentsafepath.com/registrar.html - Send as: Authorization: Bearer <key> ## Tools - search_businesses(intent, limit): natural-language business/company/local search (companies, services, hotels, restaurants, coworkings, institutions, contacts). - get_business_profile(company_id): contacts, address, website and provenance when available. ## Use cases - Find a real business with a phone/contact before acting. - Company lookup / verify a company exists, with source-backed data. - Local business search for agents (city + category + contact). - Source-backed company data instead of hallucinated details. ## Coverage - ~2.7M businesses and growing: Brazil (public registries), US (SEC EDGAR), Europe (GLEIF). ## Data policy / provenance - Licensed/public sources only. Every field carries a source URL; the index never invents phone numbers, addresses or emails. trust_score (0-100) reflects how safe an entity is to commit to. Verify provenance before acting. ## Beta & limits - Public beta. Free tier ~100 queries/day, 1500/month per agent (429 with Retry-After when exceeded). One key per agent; the key is shown once. ## Links - Website: https://agentsafepath.com - For developers: https://agentsafepath.com/developers.html - For agents: https://agentsafepath.com/agents.html - GitHub: https://github.com/saviodamato/agentrank-mcp - PyPI: https://pypi.org/project/agentrank-mcp/ - Terms: https://api.agentrank.intelligenceofcode.com/api/agent/terms