How can I parse bounce messages automatically?

Bounce messages follow a standard format called Delivery Status Notifications (DSN). When a message can't be delivered, the receiving server returns a MIME-formatted report with two machine-readable parts: the classification code and the human-readable diagnostic message. Parsing these automatically is how you build reliable suppression logic.

The structure you're working with: a bounce message is a multipart/report MIME message. Inside, there's a message/delivery-status section that contains the actual status fields:

• Status: a three-part code like 5.1.1. The first digit (5) means permanent failure. 4 means temporary. The rest classifies the failure type (address doesn't exist, mailbox full, etc.).
• Diagnostic-Code: the human-readable error message from the server ("550 5.1.1 The email account that you tried to reach does not exist").
• Final-Recipient: the address that failed.

How to implement parsing: use a MIME library for your language. In Python, the built-in email.parser handles this. In Node, mailparser works well. Extract the delivery-status section, read the Status field for classification, and run the Diagnostic-Code through pattern matching for finer classification (since different ISPs use different wording for the same failure type).

The practical reality for most senders: your ESP already does this and exposes classified bounce events via webhooks or API. SendGrid, Postmark, and most modern platforms send bounce webhook events with the type (hard/soft) already classified. Build your suppression logic off those events rather than raw MIME parsing unless you're running your own MTA.

If you are processing raw logs or need finer-grained classification than your ESP provides, the core pattern is: Status code first (5xx = permanent suppress, 4xx = retry), then diagnostic text matching for subcategories (hard bounce types like user unknown vs. domain not found vs. policy rejection each warrant different handling).