--- title: "How to let an AI agent prepare direct mail safely" description: "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." canonical: https://trysincerely.com/guides/ai-agent-direct-mail-permissions last_updated: 2026-08-31 --- > 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. # How to let an AI agent prepare direct mail safely > 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. Source: https://trysincerely.com/guides/ai-agent-direct-mail-permissions You ask an AI agent to prepare letters for 20 accounts. It can find the contacts, check the addresses, draft the copy, and set up the campaign. Before money is spent or an envelope enters the postal system, it has to stop. An authorized person reviews the actual recipients, proof, and price, then chooses whether to proceed. The agent works with that person's workspace role. It does not receive a more powerful service credential simply because automation is convenient. ## Separate preparation from release Most preparation is reversible. A contact search can be changed. So can a holdout proposal, a draft, or an unapproved design. Printing is different. It spends money, puts information in front of somebody outside the workspace, and may become impossible to recall once the provider accepts it. A short cancellation window is useful, but it is not an approval system. Put the boundary in the server. “Ask before spending” is a helpful instruction, but instructions can be misunderstood. The consequential tool itself should refuse to run until it receives a human-authenticated confirmation. ## Bind approval to exact arguments A useful confirmation answers the questions a person would ask before clicking: - Which workspace and person will perform the action? - Who will receive the mail? - Which address passed review, and what was its mailability verdict? - Which design version is shown in the proof? - What format, price, and budget are being approved? - What experiment settings will the campaign lock? - When does this approval expire? Bind the click to those exact answers. If the agent changes an address, design, format, or audience, the confirmation expires and the person reviews again. Approval for 20 letters cannot quietly become approval for 200 postcards. ## Inherit the person's authority OAuth should give the agent the same workspace and role the person already has. If that person is a member, the agent cannot act as an owner. Every read and write stays inside the workspace, and the server checks permission without relying on the agent's explanation of what it meant to do. The ordinary sending rules still apply after approval. Suppression, postal verification, [frequency caps](https://trysincerely.com/guides/direct-mail-frequency-caps), and [budgets](https://trysincerely.com/guides/direct-mail-budget-controls) can each stop the piece. A human click authorizes the requested action. It does not waive another safety check. ## Show consequences in human language Show the decision, not the plumbing. A confirmation card should use human names, recipient counts, destinations, and money. It should also say when the mail becomes impossible to recall. Raw tool names and a JSON argument dump make the person decode the system at exactly the moment they need to judge the consequence. Sincerely's MCP server is protected by OAuth and inherits the user's workspace permissions. Launches, one-off sends, approvals, suppression, cancellations, and other consequential changes wait for a human click. The [AI agent integration guide](https://trysincerely.com/integrations/mcp-agents) covers the connection and workflow. Let the agent prepare. Keep the mailbox and the money under a person's control. ## 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 set budgets and spending limits for direct mail](https://trysincerely.com/guides/direct-mail-budget-controls): Control direct mail spend with term credits, prepaid top-ups and gift funds, piece caps, approval thresholds, and a final preflight. - [How to manage direct mail suppression across campaigns](https://trysincerely.com/guides/direct-mail-suppression): Keep one durable do-not-mail decision per contact, check it at dispatch time, and cancel every queued piece that has not reached the print vendor. --- 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