InboxAPI

The Problem

AI agents can already send email through APIs like SES or SendGrid — but they can't receive, search, or reply. To give an agent a real inbox you either share your personal email account (a security nightmare) or wire up SMTP, DNS, DKIM/DMARC, and a server that listens for incoming mail. For an autonomous agent that needs to hold a back-and-forth conversation over email, that's a wall most builders never get past.

The Build

InboxAPI — gives any AI agent its own personal @inboxapi.ai email address with a full inbox: send, receive, search, reply, and forward. It connects to Claude, Gemini, Codex, or any MCP-compatible client with a single CLI command — no server to run, no SMTP to configure, no DNS to manage.

Built by Dini Labs (Sydney), it runs on proprietary email infrastructure built from scratch — not a wrapper around SES or SendGrid — with trust as the headline:

• Instant setup: one npm install + CLI command, no DNS or API-key juggling
• Native MCP integration: works with Claude, Gemini, Codex, and any MCP client
• Full email stack: inbox management and conversation threading, not just send-only
• Built-in security: account isolation, SPF/DKIM/DMARC authentication, and datamarking to defend against prompt injection hidden in incoming mail
• Abuse prevention: proof-of-work verification, recipient caps (5 external addresses), and rate limiting
• Credential-leak prevention: scans outbound messages for tokens and secrets before they're sent
• Audit trails: token-based authentication with a full activity log

Results

• Live at inboxapi.ai, free to start — no credit card required
• Real use cases: agent-to-agent dev-team coordination, monitoring and system alerts with diagnostic context, newsletter digest aggregation, and invoice/receipt processing
• Paid plans planned, but email is positioned as a fundamental agent capability rather than a premium add-on