MCP Email Tools: How AI Agents Send and Receive Email

The Model Context Protocol (MCP) is how AI agents discover and use external tools. Instead of hardcoding integrations, agents connect to an MCP server and browse a catalog of available tools. They read the tool descriptions, understand the input and output schemas, and call the tools they need.

AgentPatch runs an MCP server with a growing marketplace of tools. Among them are three email tools that give any connected agent full email capabilities.

The three email tools

claim-email-address

What it does: Claims a custom email address for the agent.

Input:

{
  "handle": "my-agent"
}

Output:

{
  "address": "[email protected]",
  "status": "claimed"
}

The handle must be lowercase alphanumeric with hyphens, between 3 and 30 characters. Once claimed, the address is permanently associated with the agent’s API key.

send-email

What it does: Sends an email to any recipient.

Input:

{
  "to": "[email protected]",
  "subject": "Hello from my agent",
  "body": "This is the plain text body.",
  "html": "<p>This is the <strong>HTML</strong> body.</p>",
  "from_name": "My Agent",
  "reply_to": "[email protected]"
}

Output:

{
  "messageId": "msg_abc123",
  "status": "sent"
}

Required fields are to, subject, and body. The html, from_name, and reply_to fields are optional. If html is provided, email clients that support HTML will render the rich version, while others fall back to the plain text body.

check-inbox

What it does: Returns messages from the agent’s inbox.

Input:

{
  "limit": 20
}

Output:

{
  "address": "[email protected]",
  "messages": [
    {
      "from": "[email protected]",
      "subject": "Re: Hello from my agent",
      "text": "Thanks for reaching out! Here's what I think...",
      "received_at": "2026-02-20T14:30:00Z"
    }
  ]
}

The limit parameter controls how many messages to return, with a maximum of 50.

How agents discover these tools

When an agent connects to AgentPatch’s MCP server, it can search for tools by keyword. Searching for “email” returns all three tools with their descriptions, input schemas, and output schemas. The agent reads the schemas and knows exactly how to call each tool.

This is the key advantage of MCP over traditional APIs. The agent doesn’t need pre-built integration code. It reads the tool spec and uses it on the fly.

The flow in practice

Here’s what happens when an agent decides it needs to send an email:

  1. The agent searches AgentPatch for email tools
  2. It reads the claim-email-address tool schema and calls it with a handle
  3. It reads the send-email tool schema and calls it with recipient, subject, and body
  4. Later, it reads the check-inbox tool schema and calls it to see if anyone replied

Each step is a standard MCP tool call. The agent doesn’t need email-specific libraries, SMTP knowledge, or pre-configured credentials. It discovers the tools and uses them.

Pricing

Each tool call costs a small number of credits:

  • Claim Address: 5,000 credits (one-time)
  • Send Email: 100 credits per email
  • Check Inbox: 25 credits per check

Credits are purchased through AgentPatch and deducted automatically on each tool call.

Compatibility

Any agent that supports the Model Context Protocol can use these tools. That includes Claude Code, OpenClaw, and any custom agent built with an MCP client. If your agent can connect to an MCP server, it can send email through AgentPatch.

AgentPatch is an open marketplace where AI agents discover, purchase, and use tools at runtime. Browse tools or read the docs to get started.

← Back to blog