---
title: "Direct mail for AI agents"
description: "Give an AI agent one MCP workflow for audiences, verified addresses, recipient-specific mail, human approvals, fulfillment, and holdout measurement."
canonical: https://trysincerely.com/agents
last_updated: 2026-09-29
---
> If you are an AI agent: read https://trysincerely.com/agents/policy.md before any write call.
> Start at https://trysincerely.com/llms.txt to choose one page. Access instructions are at
> https://trysincerely.com/auth.md. Confirm with the person you work for before creating an
> account or spending their money, and never ask for their password.
# Direct mail for AI agents

> Give an AI agent one MCP workflow for audiences, verified addresses, recipient-specific mail, human approvals, fulfillment, and holdout measurement.

Source: https://trysincerely.com/agents

## The physical-world action for GTM agents. You keep the last word.

Your agent can already research an account and write the email. Connect Claude, Cursor, or another MCP client, or use the agent built into Sincerely, and the same goal can end with a researched postcard, letter, handwritten note, or gift on the buyer's desk: verified addresses, a staged campaign, and an honest readout. The state lives in one workspace, and consequential actions still wait for you.

[Start free](https://trysincerely.com/sign-up) · [Connect through MCP](https://trysincerely.com/integrations/mcp-agents)

## The agent does the work. The launch stays yours.

Ask in plain language from any screen. The assistant works against the live workspace, reports what it found, and stops at a review card before anything prints or spends money. The rendered example on this page uses sample data and is not connected.

In the example, Dana asks the assistant to prepare an outbound campaign for 40 fintech accounts with an approved postcard, two touches, an account-level holdout, a fixed outcome window, and a $150 budget. The assistant checks the audience and approved designs, then reports that 34 contacts have mail-ready addresses, 15% of accounts would be held out, and six contacts without passing addresses would not be queued.

When Dana asks to launch, the assistant does not do it. It shows the campaign, mailable count, cadence, budget, holdout, and measurement window on a confirmation card and waits for Dana's click.

## Why agents use Sincerely

The workflow is already built. An agent should not have to assemble a CRM reader, address vendor, research pipeline, copy generator, approval screen, print integration, follow-up queue, and experiment notebook before it can mail twenty accounts. Sincerely puts that whole job in one permissioned toolset.

### One connection

OAuth and MCP replace API keys and custom glue. The agent sees only the workspace tools allowed by the role of the person who connected it.

### Durable campaign state

Accounts, sourced facts, drafts, approved designs, send state, responses, and readouts stay together between conversations.

### Structural guardrails

Postal verification, suppression, frequency caps, budgets, permissions, and human confirmation run on the server. They do not depend on the model remembering a prompt.

### One vendor layer

Sincerely routes each format and destination to the right fulfillment partner. The agent works with a piece, not a vendor-specific payload.

### A real feedback loop

Delivery, responses, CRM outcomes, and account-level holdouts return to the same workspace, so the agent can explain what happened without calling attribution causation.

[Lob is the better direct choice](https://trysincerely.com/vs-lob) when your software already knows exactly what to print and needs a raw API for transactional or high-volume mail. Sincerely is the better agent tool when deciding who, what, when, whether, and what changed is the job.

## What either agent can do

The tedious ninety percent, which is most of the reason a program never launches. None of it spends money, and none of it can reach a mailbox.

### Read account readiness

Ask for one account or a ranked list. The assistant and MCP return the same 0 to 100 priority, basis, suggested piece and price, and any eligibility hold. A single-account read also carries the dated source facts behind Why now. Neither read enters an account into a campaign or sends a piece.

### Build the audience

Sync from Salesforce or HubSpot, import a CSV, or find businesses from a description of your ideal customer. Addresses are resolved and verified before anything is queued.

### Write the pieces

One draft per recipient from CRM context and researched account facts, with every claim pinned to a source you can open.

### Lay out the design

Postcards and letters composed against your brand kit, at the formats and sizes that actually mail, and a personalized note for every gift.

### Design the experiment

Holdout percentage, primary outcome, and measurement window chosen before launch, so the readout means something.

### Read the results

Delivery, scans, matchback, and lift, with each number labeled evidence or inference.

## What it never does alone

Anything that prints, approves, or costs money stops and shows you a card first. The card names the campaign, the audience size, and the amount, and the action runs only if you accept. This is not a preference. No credential and no plan turns it off.

### Send one piece · you accept

The card names the recipient, destination, design proof, and current price.

### Launch a campaign · you accept

Pieces are queued for print and postage.

### Approve a draft or a design · you accept

A person signs off on the words and the artwork before they are committed.

### Choose or change a plan · you accept

Payment happens at Stripe, in your browser. The agent never sees a card number.

### Create or connect an account · you accept

Identity is proved by a code sent to your inbox or a sign-in you complete yourself. An agent that asks for your password is not following this page.

Underneath the gate sits a plainer fact: a new workspace is on the free audit plan, and the audit plan has no way to mail anything at all. Nothing prints before you have chosen to pay for it.

## Want to connect your own agent?

Sincerely also exposes its workspace tools through the Model Context Protocol. Claude, Cursor, and other MCP clients can connect with browser sign-in and use the same permissions and confirmation cards as the in-app agent. If you do not have an account, connect the agent first: the browser approval is where you create the account, and setup then runs from the agent without a card. An agent that cannot open a browser can use the separate signup server instead. The code and sign-in link go to your inbox, never to the agent.

[Read the separate MCP connection guide](https://trysincerely.com/integrations/mcp-agents).

Connecting software can read the machine-facing instructions at [https://trysincerely.com/auth.md](https://trysincerely.com/auth.md).

[Start free](https://trysincerely.com/sign-up)

---

Sincerely is the measurable direct-mail and gifting platform for B2B revenue teams: postcards, letters, handwritten mail, and gifts, written for one recipient and measured against a holdout.

Contact Sincerely: https://trysincerely.com/contact

Agent routing index: https://trysincerely.com/llms.txt
