MCP Email Server: Give Your Agent Email as Native Tools

This pillar explains what an MCP email server is, how Sairaph Mail's works, and how it relates to the REST API.

What an MCP email server is

The Model Context Protocol (MCP) is an open standard for connecting AI agents to external capabilities. Instead of you writing a bespoke wrapper around an email API and hand-describing each function to the model, an MCP server *advertises* its tools to the agent runtime, which discovers and calls them natively.

An MCP email server, then, is an MCP server whose tools are email operations. When it is connected, the agent can "send an email" or "read the latest inbound message" the same way it calls any other tool - the runtime handles discovery, schemas, and invocation.

First-party MCP support is now common among the leading agent-email and email-API vendors, so it is table stakes rather than a unique feature. What differentiates Sairaph Mail is that the MCP server fronts a real, EU-resident, two-way mailbox the agent owns - and is available on every plan. See the email-for-AI-agents pillar for that wider positioning, and the MCP email server glossary entry for the short definition.

How Sairaph Mail's MCP server works

• Transport: streamable-HTTP (the current MCP HTTP transport), not stdio - so it works with remote and hosted agent runtimes, not only local desktop clients.
• Endpoint: https://mail.sairaph.com/mcp/v1/mcp. The server is mounted at the root origin under /mcp, separate from the /app dashboard path.
• Auth: a mailbox Bearer key - the same scoped `sm_live_…` key you use for REST (read or read_write). One credential, both surfaces.
• Tools: the mailbox's actions - sending and reading/listing inbound - surface as MCP tools. The agent's available actions are bounded by the key's role, so a read-scoped key cannot send.

A quick reference for the endpoint and auth header:

https://mail.sairaph.com/mcp/v1/mcp
Authorization: Bearer $SAIRAPH_MAIL_KEY

Connecting a client

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "sairaph-mail": {
      "type": "streamable-http",
      "url": "https://mail.sairaph.com/mcp/v1/mcp",
      "headers": {
        "Authorization": "Bearer ${SAIRAPH_MAIL_KEY}"
      }
    }
  }
}

A remote agent runtime (YAML-style registration):

# Register the MCP server with your remote agent runtime.
# Transport: streamable-HTTP. Auth: a mailbox API key.
mcp_servers:
  - name: sairaph-mail
    transport: streamable-http
    url: https://mail.sairaph.com/mcp/v1/mcp
    headers:
      Authorization: "Bearer ${SAIRAPH_MAIL_KEY}"

Any other MCP-capable client - point it at the endpoint and pass the mailbox key:

# Any MCP-capable client: point it at the endpoint and pass
# a mailbox API key as a Bearer token.
ENDPOINT="https://mail.sairaph.com/mcp/v1/mcp"
AUTH="Authorization: Bearer $SAIRAPH_MAIL_KEY"
# Transport is streamable-HTTP; the mailbox's actions appear as tools.

The developer MCP guide has the full setup, including per-runtime notes.

MCP vs. REST: when to use which

They are two front doors to the same mailbox and the same scoped key - use whichever fits your runtime.

Because both honour the same scoped, optionally-expiring key and the same per-mailbox isolation, you can start an agent on MCP and drop to REST for a specific call without re-provisioning anything.

A worked use: inbox triage over MCP

A practical pattern: connect the MCP server to an agent, have it read inbound, classify each message, and reply - all through MCP tools. The inbox-triage walkthrough builds this end to end.

Next step: connect the MCP server - see `/developers/mcp` or the docs at https://mail.sairaph.com/docs.