Automated Thank-You Notes with AI

An AI agent that decides who deserves a thank-you note, drafts it in your voice, and drops it in the mail. Setup, prompt patterns, and quality controls.

use-casethank-youautomation

· Updated

The pattern

The agent watches for an event — a closed deal, a positive support resolution, a meeting that mattered. It fetches what it needs to know, drafts a short note in your voice, and calls send_letter. The recipient holds something physical, sent by software.

A minimal setup

  1. Wire Agent Mail in via /setup.
  2. Give the agent access to whatever data source decides who gets a note (CRM, calendar, support tool).
  3. Provide a short voice/style guide and a few example notes for in-context learning.
  4. Run on a schedule, or trigger from a webhook.

Prompt pattern

You are a thoughtful assistant who writes brief, hand-written-feeling thank-you notes (4-6 sentences). When you decide to send one, call send_letter. Don't send more than one note per recipient per 30 days. Skip recipients who don't have a complete mailing address.

Quality controls

  • Always pass from so recipients know who sent it.
  • Use list_sent_mail with recipient_filter to dedupe before sending.
  • Run a moderation step on the draft before the tool call. Agent Mail also moderates server-side.

Beyond thank-yous

The same shape works for condolence cards, congratulations on a promotion, follow-ups after a great first call, and similar one-shot, decided-by-context pieces of mail.

Wire Mailsnail into your agent

Drop this into your client's MCP config (or use /setup for one-line installs).

mcp.json
{
  "mcpServers": {
    "mailsnail": {
      "command": "npx",
      "args": [
        "-y",
        "physical-mail-mcp"
      ],
      "env": {
        "MAIL_PROVIDER": "managed",
        "MAIL_API_BASE_URL": "https://api.mailsnail.dev"
      }
    }
  }
}

See also