Agent onboarding
Point your AI agent at the Robotomail skill page for self-service email setup.
The skill page
Robotomail provides a dedicated agent skill page with machine-readable instructions that AI agents can follow to set up their own email. The page is available at:
https://robotomail.com/skillThe page includes step-by-step API instructions, code examples, error handling guidance, and information about rate limits and quotas. It's also available as plain markdown at /skill.md for agents that prefer text over HTML.
Giving your agent email access
The simplest approach is to include the skill page URL in your agent's system prompt or tool definitions:
You have access to email via Robotomail.
To set up your email, follow the instructions at:
https://robotomail.com/skill
Once set up, use the Robotomail API to send and receive email.
Base URL: https://api.robotomail.comAlternatively, pre-provision a mailbox and give the agent a scoped API key for just that mailbox. This is more secure and skips the signup step.
Pre-provisioned access
For tighter control, create the mailbox yourself and issue a scoped API key:
- Create a mailbox via the API or dashboard
- Create a scoped API key restricted to that mailbox
- Pass the key and mailbox address to your agent
The agent can then send/receive email and manage webhooks for its mailbox, but cannot create new mailboxes, access billing, or manage domains.
Build the agent’s email workflow
Choose or provision a mailbox, create credentials with the access the agent needs and complete account verification before sending. Connect your runtime using the API, SDKs, MCP or agent skill.
Receive a message, load the relevant thread, decide what the task requires and validate the proposed action. Check recipients and permissions in application code. Ask for human approval when the action is sensitive, then send a reply using the original RFC Message-ID in inReplyTo.
Persist the message ID, thread ID and action outcome outside the model. Deduplicate retried events and define a human handoff. Your application owns model selection, prompts, task state and approvals; Robotomail is the email layer, not a general agent builder.
Use receive and reply for the integration and agent email security for content and tool boundaries.