MCP

Trotta Mail speaks the Model Context Protocol, so an agent in an MCP client can use its inbox as tools.

#Connect

The server is at https://api.trotta.email/mcp. It uses Streamable HTTP and it is stateless. It takes the same bearer key as the REST API, or an assistant signs in with OAuth: see Connect ChatGPT, Claude and other assistants. A session cookie is ignored.

MCP client configuration
{
  "mcpServers": {
    "trotta-mail": {
      "url": "https://api.trotta.email/mcp",
      "headers": { "Authorization": "Bearer tm_live_..." }
    }
  }
}

Need a key first? Sign the agent up with one call.

#Tools

ToolWhat it does
create_inboxCreates an inbox and returns its address.
list_inboxesLists the inboxes the key can reach.
list_messagesLists messages in an inbox.
get_inbox_updatesWhat is new since a time, ranked, each with a one-line reason.
read_messageReads one message in the safe view.
wait_for_messageWaits for the next delivered message, with the same filters as the REST call.
send_messageSends a new message. Screened before it leaves. With send_at, schedules it.
reply_to_messageReplies in the thread. With send_at, schedules the reply.
cancel_scheduled_messageStops a scheduled message. It goes back to drafts.
create_draftLeaves a draft for a human to read and send. Works where the inbox allows drafts.
thread_contextReturns a thread as model-ready text with a trust level. Earlier messages come as their new part only.
read_attachment_textReads the text inside a PDF, Word, Excel, PowerPoint or OpenDocument file, instructions taken out.
answer_invitationAccepts, declines or tentatively accepts a calendar invitation.
list_heldLists what is waiting on a human.
scan_contentRuns any content through the guard and returns the verdict.

Every tool runs through the same checks as the REST route behind it. Results are the safe view, even with a mail:full key. Held mail is withheld. Tools take ids only: an argument that is not an id is refused before any request is made.

In a human's inbox an agent starts with drafts only: send_message and reply_to_message answer 403 agent_drafts_only, and create_draft is the way through. Login mail is sealed there: no tool returns it. See the security model.

A tool call
{
  "name": "wait_for_message",
  "arguments": { "inbox_id": "inb_7Qh2mXc4", "has_code": true, "timeout": 60 }
}

#Email sending

Each key and each connected app has a setting for the tools that send: send_message, reply_to_message and answer_invitation. Change it in the console, under API keys or Connected apps, or with PATCH /v1/keys/{id} and send_mode.

SettingWhat the client sees
Ask before every email (ask, the default)The tools are marked destructiveHint: true, so the client asks before each send.
Don't send email (off)The tools are left out, and a call to one answers sending_off. Use create_draft.
Send without asking (auto)The tools are marked destructiveHint: false, so the client may let you allow them once.

It only narrows what a connection can do. Mail is screened before it leaves in every mode, and the inbox's own rules and limits still apply. A change takes effect on the next request; a client may keep its tool list for up to a minute.