MCP Server

Give your AI agent email tools with zero code using the Model Context Protocol.

Last updated 2026-03-29

The LobsterMail MCP server gives AI agents email capabilities through the Model Context Protocol — no SDK code required. Agents get tools like create_inbox, wait_for_email, and send_email that they can call directly.

Setup#

Add LobsterMail to your agent's MCP configuration:

{
  "mcpServers": {
    "lobstermail": {
      "command": "npx",
      "args": ["@lobsterkit/lobstermail-mcp"]
    }
  }
}

That's it. The MCP server handles auto-signup and token persistence to ~/.lobstermail/token automatically. No API keys to configure.

Claude Code#

Add to your project .mcp.json or global ~/.claude.json:

{
  "mcpServers": {
    "lobstermail": {
      "command": "npx",
      "args": ["@lobsterkit/lobstermail-mcp"]
    }
  }
}

Cursor#

Add to your .cursor/mcp.json:

{
  "mcpServers": {
    "lobstermail": {
      "command": "npx",
      "args": ["@lobsterkit/lobstermail-mcp"]
    }
  }
}

Available Tools#

Tool	Description
`create_inbox`	Create a new `@lobstermail.ai` inbox. Supports smart naming with `name`/`org` for meaningful addresses.
`check_inbox`	List recent emails in an inbox. Returns sender, subject, and preview.
`wait_for_email`	Wait for an incoming email with optional filters. Uses real-time long-polling — returns within milliseconds of arrival.
`get_email`	Get a single email by ID with full body in LLM-safe format.
`send_email`	Send an email from an inbox (requires Tier 1+ verified account).
`list_inboxes`	List all active inboxes on the account.
`delete_inbox`	Soft-delete an inbox (7-day grace period).
`get_account`	View account tier, limits, and usage stats.

Example Interaction#

Once configured, your agent can use email naturally:

You: Create me an email address and wait for a verification code from Acme Corp.

The agent will:

Call create_inbox with name: "Your Name" → gets your-name@lobstermail.ai
Use the address to sign up for the service
Call wait_for_email with filter: { from: "noreply@acme.com" } → returns instantly when the email arrives
Read the verification code from the LLM-safe email body

Security#

All email content returned by MCP tools uses safeBodyForLLM() formatting — the body is wrapped in boundary markers that help LLMs distinguish untrusted email content from their own instructions. Emails flagged as injection risks are clearly marked with a warning.

The MCP server shares the same token file (~/.lobstermail/token) as the SDK. If your agent has already used the SDK, the MCP server will reuse the same account. If not, it auto-signs up on first use.