Turn Emails and Messages Into a To-Do List with AI
Half your to-dos are buried in email threads and chats. One prompt extracts every commitment you made and every request waiting on you, and creates real tasks — in Apple Reminders, Todoist, OmniFocus or Microsoft To Do. Free, local.
LMCP executes local tools on your device. Cloud AI providers and connected services can receive requested data; their terms apply.
The Tasks That Never Make It Onto a List
“I’ll send it Thursday” in an email, “can you review this?” in Teams, a client asking for a quote on WhatsApp — the tasks that slip are the ones that never left a conversation. Nobody stops mid-reply to copy the promise into Reminders, so by Friday half of what you owe people is scattered across an inbox and two chat apps, remembered only when someone follows up to ask why they never heard back.
An AI assistant that can read those apps does not have that problem. It goes through the actual thread, so it catches the commitment in your own words — not a keyword match, an understanding of who asked for what and by when — links it to the sender, and puts a date on it. The rest of this guide is the setup and the exact prompts that make that a five-minute routine instead of a mental list you carry around.
What You Will Need
- macOS 12 (Monterey) or later with Mail.app configured — Apple Silicon or Intel
- A task app — Apple Reminders works with zero setup; Microsoft To Do, Todoist and OmniFocus are also supported (more on choosing one below)
- Any MCP-compatible AI client — Claude Desktop, ChatGPT, Cursor, VS Code or any MCP client
You do not need API keys or OAuth tokens for Mail or Reminders — LMCP talks to both through native macOS APIs. Todoist is the one exception on this list: it connects with a one-time API token because it is a cloud service, not a Mac app.
Step 1: Install LMCP on Your Mac
Download LMCP and install it:
- Open the downloaded
LMCP-Installer.pkgfrom your Downloads folder - Follow the installer — it installs Local MCP to Applications for you (no dragging)
- Open Local MCP from your Applications folder — it appears in your menu bar
Takes about 30 seconds. Your AI clients are configured automatically.
On first use, macOS will ask you to grant permission for LMCP to access Mail and Reminders. Click “Allow” for both — without the Reminders grant, your AI can read the commitment but cannot actually create the task.
Step 2: Restart Your AI Client
Quit your AI client completely (Cmd+Q, not just closing the window) and reopen it. It needs to restart to discover the new MCP server.
- Claude Desktop — look for the hammer icon in the chat input
- Cursor / VS Code — MCP tools appear automatically in the composer or Copilot settings
- ChatGPT — add
https://local-mcp.com/mcpas a custom connector
Step 3: The Workflow — From Inbox to Task List
This is the part worth copying verbatim the first few times, then adapting once you have seen how your AI reads your own inbox. Four prompts cover most of a working week.
Morning triage: everything you owe, in one list
Your AI reads across all three sources in one pass and groups the results by sender, then waits for your go-ahead — you see the whole list before a single reminder gets created.
Pull the action items out of one long thread
Useful on the threads that ran for two weeks and forty replies, where the original ask is buried somewhere around message six and nobody has revisited it since.
Turn deadlines into dated reminders
The one-day buffer is deliberate — you find out you are behind with a day to fix it, not on the morning it was actually due.
The Friday close-out
This is the step that keeps the list honest. Without it, a list that only ever grows stops being trusted, and once that happens people stop checking it at all.
Getting Consistent Extractions
The four prompts above work as written, but three habits make the difference between an AI that occasionally misses something and one you actually trust to run unsupervised.
- Name the time window. “The last three days” or “this week” gives your AI a boundary to work inside. Leave it open-ended and it tends to either skim too far back or stop at the most recent handful of messages.
- Ask for the list before the write. Every prompt in this guide ends with “once I confirm” on purpose. Reviewing ten proposed tasks takes twenty seconds and catches the odd miscategorized email — a newsletter that used the word “deadline” in passing, say — before it becomes a reminder you have to delete.
- Re-run the close-out before the extraction. If you run the Friday close-out prompt first, the following week’s triage has fewer stale reminders to sort through, and your AI is less likely to propose a task that duplicates one already sitting in your list.
Apple Reminders or Microsoft To Do: Which Should Get the Task?
LMCP can create tasks in Reminders, Microsoft To Do, Todoist and OmniFocus, but they are not four interchangeable checkboxes — each one fits a different setup.
- Apple Reminders — the default. LMCP writes to it directly, there is no account to connect, it works offline, and it is where a task created mid-conversation shows up fastest. Pick this if you do not already live in another task manager.
- Microsoft To Do — LMCP does not talk to To Do’s API directly. It creates the task in Apple Reminders, and Reminders syncs it to To Do automatically once you have a Microsoft account connected (System Settings → Internet Accounts). If your team already lives in Outlook and To Do, pick this one — the task shows up in the To Do app on your phone within moments, with nothing extra for you to do.
- Todoist — connects separately, with a one-time API token. Worth it if you already run projects, labels and filters in Todoist and do not want a second, competing list living in Reminders.
- OmniFocus — works once you have the app installed, and fits a GTD-style setup with folders, projects and tags better than a flat list does.
If you are not sure, start with Reminders — it needs no setup, and if you later decide you want everything to live in To Do instead, the tasks you already created keep syncing. You are not starting the list over.
Doing This With AI vs by Hand vs Mail Rules
Extracting tasks from a messy inbox is not a new problem, and AI is not the only tool people reach for. Here is an honest comparison of the three approaches.
| Method | What it actually catches | Real limit |
|---|---|---|
| AI + LMCP | Implicit commitments in plain language — “I’ll get you that Thursday” — across email, Teams and WhatsApp in one pass, with a due date attached | Only sees the accounts your AI is connected to at that moment; it always shows you the list before creating anything, so a busy week still costs a minute of review |
| By hand | Whatever you personally notice and have the discipline to copy over — usually accurate when you do it | Costs minutes per message, so the cross-app promises — a WhatsApp reply, a Teams aside — are the ones forgotten first, because copying them out feels like more effort than the task itself |
| Mail.app rules & filters | Sender, subject or keyword matches — instant, free, deterministic | Matches patterns, not meaning: a rule can flag every email from a client as “needs action” but cannot tell which specific one actually asked for something, and it has no reach into Teams or WhatsApp at all |
In short: rules are fine for sorting mail into folders, but they were never built to read intent. Doing it by hand works fine until the volume does not. AI reading the same threads you would have read anyway is the version that scales without you doing the copying.
Local tools and data boundaries
- LMCP runs as a local server on your Mac and talks to Mail.app and Reminders through native macOS APIs
- LMCP executes local tools on your device. Cloud AI providers and connected services can receive requested data; their terms apply.
- No email or message content is stored, cached or logged by LMCP
- Preview and confirmation behavior depends on the tool, platform and AI client. Check the tool requirements and review the intended action before authorizing a real call.
LMCP executes local tools on your device. Cloud AI providers and connected services can receive requested data; their terms apply. Local execution alone does not establish GDPR or CCPA compliance. Assess the data you authorize, the selected AI provider, connected services and your organizational requirements.
Troubleshooting
The AI missed a commitment buried in a long thread
Point it at the specific thread by name or sender rather than asking it to scan everything at once — “read the full thread with Sarah about the Q3 contract” forces it through every message instead of skimming the most recent ones.
Tasks land in Reminders but never show up in Microsoft To Do
The sync runs through your Microsoft account, not through LMCP. Check System Settings > Internet Accounts and confirm a Microsoft or Exchange account is added with Reminders enabled — if it is missing, To Do has nothing to sync from.
Todoist or OmniFocus tools do not appear
Todoist needs a one-time connection with an API token before its tools are advertised. OmniFocus needs the app itself installed and running — LMCP does not create it for you.
Permission denied errors
macOS requires explicit permission for apps to control other apps. Go to System Settings > Privacy & Security > Automation and make sure LMCP has access to both Mail and Reminders. If it is not listed, download and reinstall LMCP.
The same task gets created twice
This usually means the extraction ran over an overlapping time window twice without a close-out in between. Ask your AI to check existing reminders before creating new ones — “skip anything that already has a matching reminder” — or run the Friday close-out prompt first so the list it is comparing against is up to date.
What Else Can LMCP Do?
Turning conversations into tasks is one workflow among many. The same install gives your AI 188+ tools across Calendar, Contacts, Notes, OneDrive and more — see Email with Claude, Claude with OmniFocus, or the full list of guides.
Frequently Asked Questions
Can AI really tell what is a commitment and what is just conversation?
Yes — it reads the actual sentence, not a keyword. “I will send it Thursday” and “thanks, that would be great” read very differently to a model that understands language, which is why it catches promises a keyword rule would miss without flagging routine small talk as a task.
Does it work across email, Teams and WhatsApp at the same time?
Yes, in one prompt, as long as LMCP has permission for each app. You can also scope it to just one — “only my email” or “just this WhatsApp chat” — if that is all you want scanned.
Which task app should I use, Reminders or Microsoft To Do?
Reminders if you want zero setup — LMCP writes to it directly. Microsoft To Do if your team already lives in Outlook: LMCP still creates the task in Reminders, and your connected Microsoft account syncs it to To Do automatically.
Is my email or chat content uploaded anywhere to do this?
LMCP executes local tools on your device. Cloud AI providers and connected services can receive requested data; their terms apply.
Can it also mark tasks complete when the conversation resolves them?
Yes — the Friday close-out prompt above does exactly that: it checks your open reminders against the week’s messages and marks the ones already handled, so your list reflects what is actually still open.
Ready-Made Recipes
Copy-paste workflows that build on the tools in this guide:
- Find every unanswered client request across WhatsApp, Slack, Teams & email
- Have your AI prep you for your next meeting
Related Guides
Comments
Questions and experiences welcome — comments are reviewed before they appear.