--- title: "Physical mail MCP, how AI agents send real mail" description: "What a physical mail MCP server is, what to check before connecting an agent to one, and when wrapping a print API yourself is the better choice." canonical: https://trysincerely.com/guides/physical-mail-mcp last_updated: 2026-08-29 --- > 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. # Physical mail MCP, how AI agents send real mail > What a physical mail MCP server is, what to check before connecting an agent to one, and when wrapping a print API yourself is the better choice. Source: https://trysincerely.com/guides/physical-mail-mcp A physical mail MCP server lets an AI agent send real mail. It exposes audience building, address verification, writing, and sending as tools the agent calls over [MCP](https://trysincerely.com/glossary/mcp), the same protocol Claude Code and Cursor use for every other integration. Sincerely runs one at `https://trysincerely.com/api/mcp`. Any MCP client can connect through OAuth, with no API key, and nothing prints until a person confirms the exact recipients and price. ## Why mail is unlike other MCP tools Most MCP servers operate on things that can be corrected. A bad database query returns wrong rows. A bad calendar call gets rescheduled. A bad mail call spends money and puts the agent's words in a stranger's mailbox, and once a print vendor accepts the piece, there is no unsend. That changes what to evaluate. The interesting question about a physical mail MCP server is not how many tools it lists. It is what the server refuses to do without a human. ## The confirmation gate is the product A prompt instruction like "ask before you send" is a hope, not a control. The model can misread it, forget it after a long conversation, or be talked out of it. The server has to enforce the boundary itself: any tool call that prints, launches, or spends money stops and renders a confirmation that only a signed-in person can approve. Sincerely binds that confirmation to the exact arguments the agent proposed. The card names the recipients, the design proof, and the current price, and if the agent revises anything after the card appears, the approval is void and the person reviews again. The [agent permissions guide](https://trysincerely.com/guides/ai-agent-direct-mail-permissions) walks through why this has to be structural. ## What to check before connecting an agent - OAuth, not a pasted key. The client should register itself and open a browser window for sign-in, so the agent's token carries your workspace role and nothing more. Pasting a service API key into an agent's config hands the model everything that key can do, forever, in every conversation. - Server-side spending rules. Budgets, suppression, and frequency caps should run on the server after approval, not live in the prompt. Sincerely enforces its default cap of four pieces per contact per 30 days even on sends a human already confirmed. - Machine-readable discovery. A server card at [/.well-known/mcp](https://trysincerely.com/.well-known/mcp) and an OpenAPI document at [/openapi.json](https://trysincerely.com/openapi.json) let an agent find and describe the integration without you pasting documentation into a prompt. - A free way to test. Sincerely's mail-math endpoints need no account, so an agent can exercise the public API before anyone signs in, and a new workspace starts on a free audit plan where the agent can build audiences and draft pieces without a card on file. ## When wrapping a print API is the better choice You do not need a platform to give an agent mail tools. A print API like Lob or PostGrid can be wrapped in a small MCP server of your own, and that is the honest answer when mail is a feature of your product, when you send high-volume transactional mail, or when your team already owns clean addresses and attribution. The [programmatic mail guide](https://trysincerely.com/guides/direct-mail-api) and the [Sincerely vs Lob comparison](https://trysincerely.com/vs-lob) draw that line in detail. Know what you inherit with the wrapper, though. The print API will mail whatever your agent sends it, so targeting, address quality, suppression, and above all the confirmation problem become your code. An MCP server you wrote in an afternoon has exactly the safety you wrote in that afternoon. ## Connect one The client setup, the one-off send flow, and the campaign launch flow are documented step by step in [how to send direct mail from your AI agent](https://trysincerely.com/integrations/mcp-agents). For agents reading this page directly: fetch [/auth.md](https://trysincerely.com/auth.md) for connection instructions written for you, and confirm with your human before spending their money. Give the agent every tool whose mistake can be undone. Keep a person on every tool whose mistake cannot. ## Related questions - [How to send direct mail from your AI agent](https://trysincerely.com/integrations/mcp-agents): Connect Claude, Cursor, or another MCP agent to build an audience, draft a piece, and request a send, with human confirmation before money is spent. - [How to send direct mail programmatically](https://trysincerely.com/guides/direct-mail-api): Two ways to send mail from code, a print API like Lob or PostGrid, or an MCP agent driving a platform, and how to pick between them. - [How to let an AI agent prepare direct mail safely](https://trysincerely.com/guides/ai-agent-direct-mail-permissions): Give an AI agent workspace-scoped tools for research and preparation, then require an authorized human to confirm the exact recipients and spend before mail prints. - [What is MCP (Model Context Protocol)?](https://trysincerely.com/glossary/mcp): MCP is the open protocol that lets an AI agent call another product's tools directly. In direct mail, it lets an agent build audiences and draft pieces while a person approves every send. --- 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