Compiled June 2026 from Spark's official docs, launch blog (May 13, 2026), and the open-source skills repo. This is the most direct tool we found for the CEO's backlog.
What it is
A command-line interface that exposes Spark's data (email, calendar, contacts, meeting notes) to AI agents — Claude Code, Cursor, OpenClaw, any MCP-compatible agent. It reads from Spark Desktop's local data store on a Mac, so nothing new is uploaded to a cloud. Works with every provider Spark supports — Gmail, Outlook, Exchange, iCloud, IMAP (so the CEO's Outlook/365 account is supported).
Two access tiers
| Tier | Cost | What the agent can do |
|---|---|---|
| Read | Free, all users | Search, summarize, fetch context, read threads, view calendar/contacts/notes |
| Triage | Included in Spark Pro ($199–$200/yr) | Everything in Read plus: draft/reply/forward/edit, archive/label/pin/snooze/move, assign in shared inboxes, and bulk operations across many emails and accounts |
Why it's the right tool for 34,000 unread
Readdle ships open-source "recipes" (github.com/readdle/spark-cli-skills) — and
several target exactly this problem:
- recipe-vacation-catchup — "Process a large backlog after time away with batch
processing." This is the 34K use case by name.
- recipe-inbox-by-category — process the inbox in priority order by smart category.
- recipe-inbox-zero — drive the inbox to zero, category-first.
- recipe-new-sender-review — burn down GateKeeper's new-sender queue (accept/block).
- recipe-draft-batch, recipe-multi-account-review, morning/end-of-day/weekly digests.
- Personas: Executive Assistant, Notification Hygiene (bulk-archive noisy
senders), Priority Tuning, Founder Mode.
The agent can bulk-archive/label across the whole backlog based on rules you describe in plain language ("archive everything older than 30 days from no-reply senders; label anything from a customer domain; surface anything addressed to me personally"). That's the safe way to clear 34K without 50-at-a-time clicking.
Setup (one-time)
- Mac required — Spark CLI is macOS only. Spark Desktop must be installed and running, signed into the account(s) you want to triage.
- Add the CEO's Outlook/365 account to Spark Desktop.
- In Spark: Settings → AI Agents → Spark CLI Setup, follow prompts.
- Set per-account level: Settings → AI Agents → Spark CLI Access →
read-onlyortriage(triage needs Pro). - Connect your agent (Claude Code, etc.). Install recipes from the GitHub repo.
Caveats / things to decide
- macOS only. If the CEO is on Windows, options: run it on your Mac with her account added to your Spark (delegated), or give her a Mac. Her org being on Outlook/365 is fine — it's the desktop OS that matters, not the mail provider.
- Triage = Spark Pro. Read tier is free but can't act (can't archive/label).
- Spark recommends trusted agents in a secure local environment — sensible, since the agent can act on a real inbox. Start an account at read-only, watch it work, then upgrade to triage.
- It acts on Spark's local copy — the account must be connected in Spark Desktop.
The kicker for us
This Claude Code environment already has a use-spark skill. If Spark Desktop +
CLI is set up on this Mac with the relevant account connected, I can drive the triage
directly from here — run recipe-vacation-catchup against the backlog, review what it
proposes, and bulk-clear safely.
Sources
- Getting started: https://sparkmailapp.com/help/spark-cli/getting-started-with-spark-cli
- Launch blog: https://sparkmailapp.com/blog/introducing-spark-cli
- Open-source skills/recipes: https://github.com/readdle/spark-cli-skills
- MacStories coverage: https://www.macstories.net/reviews/spark-mail-adds-a-mac-cli-and-agent-skills/