Email-Capable Crews with MultiMail

Give your CrewAI agents the ability to send, read, and reply to email — with graduated human oversight that matches each agent's role and trust level.


CrewAI is a multi-agent framework where each agent has a defined role, goal, and backstory. This role-based design maps naturally to MultiMail's oversight model: a senior agent might operate in autonomous mode while a junior researcher works in gated_send, where every outgoing email requires human approval.

MultiMail provides the email infrastructure your crew needs. Agents can send outreach emails, check inboxes for responses, manage contacts, and track conversation threads — all through the REST API or MCP server. Each action respects the oversight mode configured for that mailbox.

By combining CrewAI's task delegation with MultiMail's email tools, you can build crews that handle complex email workflows: a researcher agent finds contacts, a writer agent drafts messages, and a manager agent reviews the queue before delivery.

Built for CrewAI developers

Role-Based Oversight

Map CrewAI agent roles to MultiMail oversight modes. A trusted outreach agent can operate in monitored mode, while a new agent starts in gated_all where both sends and replies require approval.

Multi-Agent Email Workflows

CrewAI crews naturally split email tasks across agents. One agent researches contacts, another drafts emails, and a third monitors replies. MultiMail supports this by providing separate tools for each step.

Thread Tracking Across Tasks

MultiMail maintains email threads automatically. When a CrewAI task involves following up on a previous email, the agent can retrieve the full thread context with get_thread and craft a relevant reply.

Contact Management for Outreach

CrewAI outreach crews can use MultiMail's search_contacts and add_contact tools to build and query contact databases, tracking who has been contacted and their response history.

Mailbox-Per-Agent Architecture

Assign different mailboxes to different crew agents. Your sales agent sends from sales@, support agent from help@, and each mailbox can have its own oversight mode and sending limits.


Try it with your agent

No code, no dashboard. Paste this to your AI agent — it connects MultiMail, creates an inbox, and builds the flow for you.

1. Get MultiMail ready: read https://multimail.dev/llms.txt, follow it to connect the MultiMail MCP server, create a free inbox, and set up a verified sending domain. 2. Wire MultiMail into this CrewAI project using CrewAI’s real MCP tool mechanism: if the project uses CrewAI’s YAML/DSL style, add the MultiMail MCP server through the agent’s mcps field; if it uses Python-defined agents, use crewai-tools MCPServerAdapter and pass the loaded MultiMail tools into the Agent tools list. 3. Give the CrewAI agent only these email abilities at first: check the inbox, compose a draft reply, and send email. Keep the tool names and arguments exactly as exposed by the connected MultiMail MCP server. 4. Send one test email from the verified sender to a developer-controlled recipient, then check the inbox and draft a reply so I can confirm read, draft, and send all work end to end. 5. Run MultiMail in gated_send oversight mode for this setup, so every outbound email requires developer review before sending. Do not switch to monitored or autonomous until I explicitly ask.

Step by step

1

Create a MultiMail Account

Sign up at multimail.dev and create one or more mailboxes for your crew. Generate an API key from the dashboard.

2

Install CrewAI and Dependencies

Install CrewAI and the requests library for API calls.

3

Define Email Tools

Create @tool decorated functions for send_email, check_inbox, reply_email, and any other MultiMail operations your crew needs.

4

Design Your Crew

Create agents with roles that map to email responsibilities. Assign each agent only the tools it needs — a reader agent does not need send_email.

5

Run and Review

Kick off the crew with crew.kickoff(). If using gated_send mode, review and approve pending emails in the MultiMail dashboard.


Common questions

Can different CrewAI agents use different oversight modes?
Yes. Each MultiMail mailbox has its own oversight mode. Assign different mailbox IDs to different agents — a trusted agent uses a mailbox in monitored mode, while a new agent uses one in gated_all. This gives you fine-grained control over which agents can send freely.
How do I prevent a crew agent from sending spam?
Use gated_send mode (the default) so every outgoing email requires human approval before delivery. Additionally, you can set sending limits on your MultiMail plan and use the agent's backstory to instruct it on acceptable email volume and content standards.
Can CrewAI agents handle email attachments?
Yes. The MultiMail API supports attachments on send and reply endpoints. Create a tool that accepts file paths, encodes the content as base64, and includes it in the API request body. Your agent can then attach files as part of its email workflow.
How does CrewAI's sequential task execution work with email?
CrewAI executes tasks in order, passing output from one task to the next. This works well for email workflows: a research task outputs contact information, which feeds into an outreach task that drafts and sends emails. Each task's output becomes context for the next agent.
What happens if the MultiMail API returns an error?
The tool function returns the error response, and the CrewAI agent can reason about it. Common errors include rate limits (429), authentication failures (401), and validation errors (400). Your agent can retry, adjust its approach, or report the issue based on the error type.
Can I use CrewAI's memory features with MultiMail?
Yes. CrewAI's memory system helps agents remember past interactions. Combined with MultiMail's thread tracking, your agents can maintain context across email conversations over multiple crew runs, recalling who they have contacted and what was discussed.

Explore more

The only agent email with a verifiable sender

Email infrastructure built for AI agents. Verifiable identity, graduated oversight, and a hosted MCP server. Formally verified in Lean 4.