Mail Inspector

Inspect the legitimacy of the emails you receive in Apple Mail on macOS

Overview

Mail Inspector is a native macOS app for inspecting potentially malicious email without opening it in Apple Mail. Drop an .eml or .emlx file, open one or more from Finder, or drag a message onto the Dock icon, and Mail Inspector parses the raw RFC 5322 source and builds a report with the sections below.

Mail Inspector never renders HTML, never executes scripts or attachments, and never follows links. Message content is never logged or persisted beyond what you explicitly import.

Report sections

• Noteworthy Observations (off by default) — an optional summary at the top of the report, gathering every flagged finding from the sections below into one list. Severity labels and plain descriptions only, deliberately not a single score, so you form your own judgment rather than anchoring on one number.
• Sender identity — the actual From address and display name, with discrepancies (forged display names, Reply-To/Return-Path mismatches, IDN/Punycode domains, mixed-script display names) surfaced as observations, not accusations. A Reply-To mismatch you've reviewed can be marked safe for that sender from right there in the report, so it stops being flagged. Can optionally show the sender domain's published BIMI brand logo, but only once DMARC is a trusted pass (and, optionally, only above/below a spam-score threshold you set).
• Authentication — SPF, DKIM, and DMARC results parsed from Authentication-Results and DKIM-Signature headers (including full DKIM signature detail: algorithm, canonicalization, signed headers, selector, signing identity, timestamps), with an explicit trust model: a result is only shown as trusted when it comes from an authserv-id you've configured (or, by default, treated as trusted outright — configurable in Settings). DMARC domain-alignment (strict/relaxed) is shown for both SPF and DKIM. An optional, manually-triggered SPF recheck can re-run the check against the record as it exists right now via a real DNS lookup — one of only two network requests this app ever makes.
• Spam Likelihood (off by default) — a gauge built from a spam-scoring header your mail provider's filter already added (X-Spam-Score, Exchange's Spam Confidence Level, Rspamd, mailbox.org, etc.), rescaled for display. Never an independent assessment — the report always says so.
• Delivery path — a reconstructed, chronological timeline from every Received header, with IP-scope classification (private/loopback/reserved), reverse-DNS hostname mismatches, TLS version/cipher per hop, and chronological/timing anomalies flagged — all without treating anything outside your own trusted infrastructure as proof of wrongdoing. A hostname mismatch you've reviewed can likewise be marked safe for that specific claimed/verified pair.
• AI Message Analysis (off by default) — a 0-100 legitimacy gauge, a short prose analysis, and a follow-up chat, generated from the signals already shown elsewhere in the report (SPF/DKIM/ DMARC, delivery path, spam score) — never the raw headers or message body, unless you explicitly opt in to attaching a text excerpt to a specific chat question. Runs fully on-device via Apple Intelligence (macOS 27+, eligible hardware, zero configuration), fully on-device via a model you download once and run locally through MLX (Apple Silicon Macs only — a small, broadly-compatible model and a larger one gated at 16GB of memory), or against a remote provider you configure: LM Studio or another local OpenAI-compatible server, and a catalogue of hosted providers (OpenAI, Anthropic, Google, Mistral, Cohere, DeepSeek, Groq, MiniMax, OpenRouter, Perplexity, Scaleway, GitHub Models), plus a fully custom OpenAI-compatible endpoint. Two further options, Jev and System One (Jev) compatible, use TypeSafe.ai's System One API instead — a fast decision-maker that returns only a 0-100% legitimacy score for each message, with no further insight and no chat against the message's headers or content. Jev talks to TypeSafe.ai's own hosted service; System One compatible points at any live or locally-hosted service speaking the same API (e.g. Laya (local), Clef (local, online)) — the built-in prompt is tuned for TypeSafe.ai's Jev and may not give satisfactory results with other services. API keys are stored in the Keychain, never in plain settings. It's a model's opinion, not a verdict — it can be confidently wrong. The system prompt sent to whichever chat-capable provider is active (and, for Jev/System One compatible, the question it asks instead) is itself editable in Settings, with a "Reset Prompt" button to restore the tested default if an edit makes things worse.
• Additional Filtering Headers — known chain-of-custody and anti-spam headers this app doesn't otherwise parse structurally (ARC-, Received-SPF, X-Spam-, Microsoft 365/Exchange anti-spam headers, Rspamd, mailbox.org), shown as-is with a plain-language explanation of what each means. Never reimplements any filter's scoring logic.
• Raw headers — a searchable, monospaced, read-only view of the complete original headers, with one-click copy of a single header or the entire header block.

Report export

A message's report can be exported or shared as a paginated PDF (⌘E to save, ⇧⌘E to share via the system share sheet — Mail, Messages, AirDrop, Save to Files, …), at a page size you choose (US Letter, US Legal, A4, or A5) in Settings. If the AI analysis has already been generated for that message, its score and analysis are included in the export.

Settings

• Trust configuration: which authserv-ids to trust for Authentication-Results (or trust all by default), plus the Reply-To-domain and delivery-hop-hostname exceptions you've marked safe from the report itself.
• Import limit: maximum accepted message size, checked before and after reading untrusted input.
• Domain alignment data: whether to keep a weekly-refreshed copy of the public suffix list (used to tell a subdomain apart from an unrelated domain for DMARC alignment) up to date — the only other network request this app ever makes, besides the manual SPF recheck above. Off by default falls back to a small bundled list.
• Noteworthy Observations, Spam Filtering, Brand Images (BIMI), and AI Message Analysis: each toggled independently, all off by default except where noted above.
• Report export page size.

Important

You cannot drag a mail directly onto the app window. This is a limitation of macOS and Mail.app. When dragging an email onto an application, Mail.app only includes a message ID to create a deep link which opens the mail in Mail.app. It does not send the mail content itself, let alone the mail headers we actually need to analyse. When you are dragging an email onto an app's Dock icon, though, it sends the entire email message – and that's why that drag and drop operation works. Yes, it is annoying and inconsistent. That's how Apple designed it. I can only work with what Apple gives me to work with, folks.

Requirements

• macOS (latest SDK), Swift 6, SwiftUI + AppKit.
• Third-party dependencies: mlx-swift-lm and swift-transformers, used only by AI Message Analysis's on-device MLX option.
• AI Message Analysis's Apple Intelligence option requires macOS 27+ with Apple Intelligence enabled on eligible hardware; its on-device MLX option requires an Apple Silicon Mac with enough free memory for the model you pick. Every other feature, including the rest of AI Message Analysis (a remote provider), works without either.

Building

Open mailinspector.xcodeproj in Xcode and build the Mail Inspector scheme.

License

Mail Inspector is released under the MIT License. See license.txt for the full text.

Copyright (c) 2026 Nicholas K. Dionysopoulos / Akeeba Ltd.