Postfleet is an inbound email infrastructure for AI agents. It operates on the premise that any message could potentially be an attack.
Every message undergoes a uniform process before it reaches your agent. This includes SPF/DKIM/DMARC checks, HTML-to-text cleaning, removal of hidden elements, HTML comments and zero-width characters, malware and spam screening, and finally, a prompt-injection scan. The result is a typed JSON that matches your defined schema.
Postfleet has a failsafe mechanism. Malware is quarantined, spam is filtered out, and a message flagged for injection risk never reaches extraction. Instead, your agent receives an explicit status (skipped_injection_risk), preventing any quietly poisoned context. The pre-sanitization body is never sent to agent-facing tools.
You can connect over MCP with 16 tools (Claude, Claude Code, Cursor, Windsurf, Cline, VS Code) or use the REST API. Sends are idempotent, meaning if you retry with the same client_id, the email is sent out only once. There is also an option for human approval before any irreversible action is taken.
We maintain transparency by publishing the red-team scoreboard. This includes details of what got through and one false flag we caught ourselves. We've successfully caught 99.1% of 113 attacks, most of which were written by independent models and a public benchmark, rather than by us. These were screened against known patterns, and we never claimed immunity.
Our service is free for up to 100 messages and 3 mailboxes. Sanitization and injection screening are always free and are not metered.
Thestral12 