The short answer
ChatGPT can help draft and check a business letter. To mail it through IRIS, either upload a reviewed PDF on the website or build a trusted application that connects an OpenAI model to IRIS REST or server MCP. Typing a request in an ordinary chat does not, by itself, create a mailing.
Choose the right path for your letter
For an occasional customer letter, start with the website. Give ChatGPT the facts and tone you want, check the draft against your source records, export it to a PDF, and use IRIS to review and mail the file. You do not need to build an integration to use AI during preparation.
For a recurring business process, a developer can build a small application around the IRIS API. The application handles credentials, document storage, review, and order submission. The model helps with bounded tasks such as drafting from approved facts or explaining a quote. These are custom integration patterns, not a claim that IRIS is a preinstalled ChatGPT app.
Ask for a draft that is easy to verify
Provide the purpose, approved facts, recipient context, and desired next action. Ask the assistant to flag missing information instead of guessing. Keep identifiers and confidential material out of the prompt when they are unnecessary for drafting. A model does not know whether your billing record or contact address changed this morning.
For example, a service confirmation can state the completed work, the date, a support contact, and the next step. Ask for a concise draft using only those inputs. Verify names, dates, prices, and commitments before converting it to a document. Inspect the exported PDF as well; correct chat text can still become a badly paginated letter.
Draft a one-page customer service confirmation using only
the approved facts below. Use plain language and one next step.
Mark missing facts as questions for me. Do not invent dates,
addresses, prices, guarantees, or a signature.
Return a draft for review; do not send anything.Connect an OpenAI application to IRIS
OpenAI’s Responses API supports remote MCP tools and an approval exchange. An application can restrict the exposed tools and handle approval requests before execution. The sample configuration below permits only an account-balance lookup; it is a configuration fragment for your server application, not a complete chat app or a mailing command.
Supply the IRIS bearer token from your application’s secret storage. A remote MCP connection lets the model provider call IRIS on your account’s behalf, so use a suitably scoped credential and send only the data needed for the task. The application must implement the approval exchange; require_approval is not a substitute for a review screen showing the final PDF.
{
"type": "mcp",
"server_label": "iris",
"server_url": "https://api.irispostalforms.com/mcp",
"authorization": "SERVER_SUPPLIED_IRIS_API_KEY",
"allowed_tools": ["get-account-balance"],
"require_approval": "always"
}Add a reviewed mailing step to your application
Build preparation first: private PDF upload, document registration, and quote. Your server can handle those operations directly through REST while the assistant explains the resulting review summary. Do not put the IRIS API key into browser JavaScript or ask a user to paste it into a conversation.
Only expose the submission action when a person has approved the exact document, sender, recipient, options, price, and purpose. Save one idempotency key per mailing, then preserve the order ID after submission. If you use Codex to implement this workflow, ask it to create preparation and review code first, with mocked responses for tests; coding assistance is separate from permission to send a real letter.
Know what a successful result means
The result you want is a reviewed order linked to the document your team intended to send. A draft, an upload URL, a quote, and a paid order are separate milestones. None should be described as “delivered.” Read the IRIS order status for progress and available tracking.
Start with one routine, non-sensitive business letter. Once that workflow is understandable to the person reviewing it, add the exception handling your team needs: address corrections, insufficient credits, rejected documents, and uncertain submission responses.
Sources & further reading
Product details checked against the IRIS API contract when this article was published. Platform references can change; consult the linked documentation when implementing.
Published by IRIS Postal Forms, operated by Great Lakes Computing Office LLC. Send a correction.