For AI agent builders

MCP-native email for AI agents

Give your agent its own inbox. Receive, parse, and reply to email via MCP. Human supervision built in. Remote server, no local setup.

Get your API key — free Read the docs

How it works

Three steps to an agent inbox

1

Create an agent inbox

In the dashboard, create a new agent inbox. You get a domain, API key, and MCP endpoint instantly.
2

Connect your agent

Add the MailKite MCP server to your agent's config. One JSON block and your agent has an inbox.
3

Agent receives and acts

Email arrives, the agent reads it via MCP, and can reply or forward. You supervise from the dashboard.

Show me

One config block and your agent has an inbox

Add the MCP server to your agent's config. That's it.

json

// Connect Claude Code to MailKite via MCP
// 1. Add the MailKite MCP server to your config:
{
  "mcpServers": {
    "mailkite": {
      "command": "npx",
      "args": ["@mailkite/mcp"],
      "env": {
        "MAILKITE_API_KEY": "your-api-key"
      }
    }
  }
}

// 2. Your agent now has these tools:
// - mailkite_receive  — read incoming email
// - mailkite_send     — send email from the agent's inbox
// - mailkite_list     — list messages in the inbox
// - mailkite_reply    — reply to a specific thread

Why teams choose MailKite

What you get

Remote MCP server

Connect any MCP-compatible agent to MailKite's hosted MCP server at mcp.mailkite.dev. No local setup required.

Claude Code plugin

First-class plugin for Claude Code — install once, your agent gets an inbox automatically. Works with Claude Desktop too.

Per-agent inboxes

Every agent gets its own domain, inbox, and API key. Isolate agents from each other and from your production mail.

Human-in-the-loop

Agents can receive freely. Outbound sends require explicit approval or pre-authorized templates. Full audit trail.

Two-way threads

Agents can follow up on conversations. threadId links replies to the original message — the agent maintains context.

Scope limits and rate caps

Set per-agent limits on send rate, recipient domains, and message size. The agent can't exceed its bounds.

In the box

What's included

Remote MCP server at mcp.mailkite.dev
Native Claude Code plugin
Per-agent domains and API keys
Human-in-the-loop approval for outbound
Two-way email threads with context
Scope limits and send-rate caps
Full audit trail for every agent action
OAuth for agent authentication

Questions

What is MCP email?

MCP email is email integration for AI agents via the Model Context Protocol. MailKite provides a remote MCP server that gives any Claude-compatible agent its own inbox — receive, parse, and reply to email with tool calls.

How does an agent receive email?

Each agent gets its own domain and webhook. When email arrives, MailKite parses it and the agent can read it via MCP tools. The agent can also send replies — all from within the MCP conversation.

Is this safe? Can agents send email without supervision?

MailKite has built-in guardrails: per-agent scope limits, send-rate caps, human-in-the-loop approval for outbound, and full audit logs. Agents can receive freely but outbound requires explicit approval or pre-authorized templates.

Which agents are supported?

Any agent that speaks MCP: Claude Code, Cursor, Windsurf, VS Code with Copilot, and custom agents built on the Agents SDK or MCP SDK. MailKite also has a native Claude Code plugin.