Reachant

• Tenant isolation derived from the logged-in user: every read goes through `scoped(Model)`, which raises if no account is set. A route missing its guard fails loudly instead of leaking another account's data.
• Two enrichment pipelines, picked per search by market. Brazil: Firecrawl for website and CNPJ (company tax ID), then BrasilAPI for legal name and partners. US: Perplexity and Serper for decision-makers. Leads are enriched in parallel with a configurable concurrency cap.
• Credits are debited with a conditional `UPDATE`, so concurrent workers can't overspend. With no balance left, the paid step is skipped and the lead keeps its basic data. Each debit logs the vendor cost to a ledger.
• Idempotent Stripe webhook: the signature is verified and every `event_id` is stored, so a replayed event or a race between workers is a no-op.
• Searches run off the request in a thread pool, and all scraping goes through a `safe_get` that blocks private IPs and re-checks every redirect (SSRF protection).
• Two-step v1 API for agents: a free estimate returns a signed token, and search creation is idempotent on that token. An MCP package exposes it to clients like Claude.

• 9 tools in the MCP server.
• 558 pytest test functions across 82 files, plus 17 in the MCP package.
• 4,036 business categories for the US and 3,872 for Brazil.
• UI in Portuguese and English.

Shows I can build a SaaS that spends money on third-party APIs without losing control of cost, with billing, tenant isolation and access for AI agents.