Record Discord Calls on Mac: Local Transcript, No Bot

Discord stopped being just for gaming years ago. Podcast interviews get taped there, communities run town halls there, startups do user research there, and half the freelance world takes client calls there because the client was already on the server. Then the call ends and it’s gone — Discord has no record button, on any plan. If you’ve tried to record Discord calls on a Mac, you’ve likely met the two standard answers: a recording bot that only works in servers you control, or a screen recorder that gives you a video file and no transcript. Neither is the actual job.

The actual job is a transcript you can search and a summary you can share. On a Mac, both can be produced locally from the audio the Discord app is already playing — without a bot, without touching the server, and without the call leaving your machine. This post covers what Discord offers natively (nothing), what the bot route can and can’t do, and the local workflow end to end.

What Discord gives you natively

The precise answer is quick: Discord has no built-in call recording. Not in voice channels, not in DM calls, not in group DMs. Nitro adds streaming quality and upload size, not recording. There’s no transcription feature either — no live captions of calls, no post-call notes. When a call ends, the audio is gone, and anything decided in it lives in the participants’ memories and whatever someone typed into chat.

This is unlikely to change soon, and the reason matters for what comes next: Discord’s culture leans hard on ephemerality and privacy expectations. Voice is treated as off-the-record by default. Any recording you make, by any method, comes with the obligation to say so — more on that below.

The bot route: what Craig does well, and where it stops

Craig is the recording bot everyone recommends, and within its lane it’s genuinely good. Invited into a server’s voice channel, it records multi-track audio — each speaker as a separate file, which is a real gift for podcast editing — for up to six hours per session, and it announces itself so everyone knows recording is happening.

The lane is narrow, though:

• Servers only. Craig must be added to a server by someone with permission to manage bots. It cannot join DM calls or group DMs at all — the conversation types where interviews and client calls usually happen.
• You need standing in the server. Recording a community call in someone else’s server means asking an admin to add a bot and grant it access. For a one-off interview in a server you don’t run, that’s a big ask.
• It produces audio files, not notes. Craig hands you tracks to download — transcription and summarization are your problem afterwards, with the download link expiring after about a week.
• It’s a third party in the call. The audio routes through the bot’s infrastructure. For a sensitive conversation, that’s another place the recording exists.

If you’re producing a podcast from a voice channel in your own server and want per-speaker tracks for editing, use Craig — that’s the use case it was built for. For everything else, especially DM calls, the bot route is structurally closed, which is the same wall that bot notetakers hit with Slack huddles: no join mechanism, no bot.

The local path: capture what the Mac is already playing

Every word of a Discord call passes through your Mac twice — everyone else’s voice through Discord’s audio output, yours through the microphone. Two macOS frameworks turn that into a recording pipeline, the same pair behind the no-bot approach to system-audio capture:

• ScreenCaptureKit with an audio-only filter captures the output of a specific process — here, Discord — while everything else on the Mac stays out of the recording. macOS groups the permission under Screen & System Audio Recording, but it’s audio only: never video, never a pixel of the screen.
• SpeechAnalyzer, Apple’s on-device speech framework, transcribes the stream live as the call runs. No audio leaves the Mac at any point, and once the OS has fetched the language model, transcription runs fully offline.

Dictanta connects the two, then generates a structured summary on-device with Apple Foundation Models when you stop: a TL;DR, action items with best-effort owner and due-date inference, decisions, and open questions — every bullet audio-anchored, so clicking it plays the moment it came from. For an hour-long community call, the decisions list with checkable audio behind it is the artifact you actually wanted.

One Discord-specific detail: Discord’s desktop app isn’t one of the apps Dictanta auto-detects (those are Zoom, Teams, Webex, Slack, and the major browsers). That changes exactly one click in the workflow, covered below. If you take Discord calls in Chrome, Safari, or another major browser instead, the browser is auto-detected and the call shows up as the This meeting option on its own.

How to record Discord calls on a Mac, step by step

Assume Dictanta is installed on macOS 26.4 or later on Apple silicon.

1. Join the call and bring Discord to the front

Voice channel, DM call, group DM — the capture doesn’t care, because it never interacts with Discord’s servers. Make sure the Discord app is the frontmost window for the next step.

2. Start recording with the “Another app” source

Click Dictanta’s waveform icon in the menu bar, hit Record, and pick Another app from the source picker. That option captures the frontmost app’s audio — Discord — with your microphone mixed in automatically: their half of the call arrives through Discord’s output, your half through the mic, and together that’s the whole conversation. Once recording starts, the capture is scoped to Discord’s process, so you can switch to your notes app mid-call without losing the audio.

The first system-audio recording triggers a one-time macOS permission prompt under Screen & System Audio Recording. Grant it once. Discord calls often start on impulse — someone hops into the channel and rings you — so the optional global shortcut is worth enabling: flip on “Global record shortcut” in Settings and ⌃⌥⌘R starts a recording from anywhere. Just bring Discord frontmost first, since that’s what Another app points at.

3. Say you’re recording

Craig announces itself when it joins a channel; a local recording announces nothing, so the announcing is on you — and in Discord’s voice culture, doubly so. “I’m recording this for notes, it stays on my machine” takes three seconds. Recording consent law varies by jurisdiction (one-party vs. all-party consent) and applies to a local capture exactly as it does to a bot.

4. Stop, and read the notes

When the call wraps, stop from the menu-bar panel or the global shortcut. The transcript is already complete — it streamed live during the call, with per-segment timestamps you can click to jump the audio. The summary generates on-device right after, and the recording titles itself with a timestamp before auto-renaming from the summary topic. The transcript autosaves every 30 seconds during the call, so if Discord (or the Mac) dies mid-call, you get a recording marked “Recovered” rather than nothing.

5. Share it where the community lives

Dictanta exports Markdown, PDF, Word (.docx), SRT, and JSON, and the Share Recap action produces a clean summary message. For a community call, that’s a recap pasted into the announcements channel; for a podcast interview, the SRT export feeds your subtitle workflow directly.

Discord-specific things worth knowing

It works where the bots can’t. DM calls and group DMs are recorded exactly like server voice channels, because the capture happens on your Mac, not in Discord. The conversations that matter most — interviews, client calls, one-on-ones — tend to be precisely the ones Craig can’t reach.

Everything Discord plays is in the recording. Screen-share audio, a music bot in the channel, soundboard effects — if it comes out of the Discord app, it lands in the recording and the transcript picks up whatever speech it contains. Mute what you don’t want captured.

Server audio is one mixed stream. Unlike Craig’s per-speaker tracks, local capture receives the channel’s audio as one mix, which cascades into the transcript: no speaker labels (diarization is planned for Dictanta, but it isn’t shipped). For interview transcription, two voices are easy to tell apart on the page; for a rowdy eight-person channel, expect to lean on the audio anchors to check who said what.

Search makes old calls useful. Dictanta’s search covers titles, transcripts, and summaries on-device, so “what did we decide about the moderator rotation” is findable weeks later — the thing a pile of Craig audio files never gives you without transcribing them yourself.

Where the local path has limits

The honest list:

• No per-speaker tracks. If you’re editing a podcast and need each voice isolated, Craig’s multi-track output wins and this approach doesn’t substitute for it. The local path optimizes for transcripts and notes, not stems.
• No speaker labels in the transcript. Same root cause: one mixed stream. Planned, not shipped.
• Mac only. Take the call on your phone or iPad and there’s no Mac audio to capture — system-audio capture is Mac-exclusive. (Headphones are fine, though: the output device never matters.)
• It records from the button press. No retroactive capture. The global shortcut exists to shrink the gap between “call started” and “recording started.”

When to use which

Use Craig when you run the server, everyone expects recording, and you need per-speaker audio tracks for production work. Use local capture when the call is a DM or group DM, when you don’t control the server, when you want a transcript and summary rather than raw audio, or when the conversation shouldn’t route through anyone’s infrastructure — the same no-bot logic that applies to every meeting platform, applied to the one platform that never had a record button to begin with.

Trying it on your next call

1. Install Dictanta from the Mac App Store.
2. Grant the microphone and system-audio permissions, and enable the global record shortcut in Settings.
3. On your next Discord call, bring Discord frontmost, press ⌃⌥⌘R, pick Another app, and say you’re recording.
4. When it ends, read the summary and click any bullet to hear the moment behind it.

The first 3 completed recordings are free with every feature unlocked and no length cap — a two-hour community call counts the same as a five-minute check-in. After that it’s $9.99/month, $79.99/year, or $149.99 once for lifetime. Transcripts and summaries sync to your other Apple devices through your own iCloud; the audio never leaves the Mac.

Bottom line

Discord gives you no recorder, and the bot everyone recommends can’t enter the DM calls where your important conversations actually happen. Your Mac was always capable of capturing the call it’s playing: scoped system-audio capture plus on-device transcription turns any Discord call — server channel, DM, or group DM — into a searchable transcript and an audio-anchored summary, with nothing joining the call and nothing uploaded anywhere. Announce that you’re recording, press the shortcut, and the hour-long call that would have evaporated becomes notes you can actually use.

FAQ

Can you record a Discord call without a bot?

Yes, on a Mac. Discord itself has no recording feature, but macOS can capture the audio the Discord app is playing plus your microphone, which together are the complete call. Nothing joins the call, nothing is uploaded, and it works in DMs and group calls — the places recording bots can’t go at all.

Can Craig record Discord DM calls?

No. Craig only works in server voice channels — it has to be invited to a server by someone with permission to add bots, and it cannot join direct-message or group DM calls on any platform. If the conversation you need to record happens in a DM, the bot path is closed and local capture on your own machine is the workable option.

Does Discord notify people when you record a call this way?

No, because nothing happens inside Discord — the recording captures your Mac’s audio, not the call itself. That makes disclosure your responsibility. Say you’re recording at the start of the call, the way Craig announces itself in a voice channel. Consent law for recorded conversations varies by jurisdiction and applies regardless of the tool.