What is Molted?
Molted is email infrastructure for developers running AI agents that need their own inboxes. It screens incoming mail for prompt injection before an agent reads it, checks outgoing messages against sending rules, and can hand risky or difficult conversations to a person. Agents connect through MCP, a CLI, or REST, with support for tools including Claude Code, Cursor, and Codex.
Why Molted works
An email agent can treat instructions written by a stranger as part of its job, which makes an ordinary inbox a dangerous place to give autonomous software access. Molted sits between the agent and every message, quarantining suspicious inbound content and checking outbound actions before they leave. It makes the agent work through guarded email infrastructure instead of trusting whatever arrives in its inbox.
Molted Email features
- Prompt injection quarantine. Incoming email is screened before the agent reads it. High risk messages are quarantined so the agent sees that content was held, while only a person can release the original.
- Policy checks on every send. Outgoing mail is checked for duplicates, cooldowns, suppressions, limits, and other mailbox rules before delivery.
- Human handoff. Agents can hand a conversation to a person with a note. While the person owns the thread, the agent cannot email that contact until the thread is handed back.
- Adjustable autonomy. Each mailbox can run autonomously or require approval for first contact or every outgoing email. Unanswered approval requests expire after 24 hours without sending.
- Agent friendly access. The same inbox can be used through a hosted MCP server, the Molted CLI, or a REST API, so agents can read, send, and react to email programmatically.
- Multi agent coordination. Several agents can share one mailbox. Contact leases and shared duplicate checks help stop two agents from emailing the same person at once.
Who Molted Email is for
- Developers giving Claude Code, Cursor, Codex, or custom agents access to email and needing checks between untrusted messages and agent actions.
- Support teams running agents that answer customer email but need risky requests or sensitive conversations handed to a person.
- Sales automation teams letting agents contact prospects while requiring approval for first contact or tighter rules around repeated sends.
- Indie developers and freelancers who want an agent owned inbox on their own domain without setting up a separate email delivery provider.
Similar micro SaaS ideas you can build
- Approval layer for AI finance agents. A control service for agents that can issue refunds, update billing records, or make other sensitive account changes. Low risk actions pass automatically, while unusual or high impact actions wait for a person before execution.
- Trust gateway for agents reading external documents. A screening layer for AI agents that ingest customer uploads, shared documents, or form submissions. It isolates instructions hidden inside untrusted content so the agent can extract useful data without treating every sentence as an order.
- Contact ownership for autonomous sales agents. A coordination tool for companies running several outbound agents against the same prospect list. Each agent temporarily claims a contact before acting, which stops duplicate outreach and conflicting follow ups without forcing all agents through one queue.
