Beyond “Summarize My Inbox”

MCP (Model Context Protocol) lets AI assistants take real actions on your Mac. The first thing everyone tries is “summarize my inbox” — but the productivity payoff comes from the next step: triage it, file what's done, draft the replies, and put the follow-ups on your calendar, in one prompt.

LMCP is the most comprehensive MCP server for macOS: 183 tools that read and act across your apps, all running locally with no API keys or OAuth. This guide is organized around the jobs people actually do (based on anonymized, aggregated fleet usage), with the tools and prompts for each.

Job 1 — Inbox & triage (the #1 use)

Email is the single most-used job on LMCP, and the standout real pattern isn't reading — it's reading and then bulk-filing: sorting messages into folders, replying to the rest.

Read: list_emails, read_email, search_emails (across all accounts)
Act: send_email, reply_email, move_email (file into folders), save_attachment, create_email_folder

Go through my unread inbox. File newsletters into “Read later,” draft replies to anything that asks me a direct question, and list anything that needs a decision so I can handle it myself.

Job 2 — Plan & schedule

The most reliable action jobs on the fleet are scheduling and documents — they almost never break mid-flow. Your AI reads your calendar across iCloud, Google, and Exchange, then creates and moves events.

Read: list_calendar_events, list_calendar_names
Act: create_calendar_event, update_calendar_event (reschedule), delete_calendar_event, plus create_reminder / complete_reminder and OmniFocus / Microsoft To Do tasks

Find a free 30-minute slot tomorrow afternoon, schedule a “Contract review” with Jana, and add a reminder to prep the day before.

Job 3 — Team catch-up & messaging

Read what was decided in Teams, Slack, WhatsApp, and iMessage — and reply where you can. (Messaging is the job that breaks most often, usually a permissions or app-state gap; run run_diagnostics if a chat tool fails.)

Read: teams_read_chat_messages, teams_read_channel_messages, slack_read_channel_messages, slack_search_messages, whatsapp_read_messages, read_messages (iMessage)
Act: teams_send_message, teams_send_channel_message, whatsapp_send_message, whatsapp_send_file

Teams works with no Graph API and no Azure AD — LMCP reads the local Teams cache directly.

Catch me up on the #engineering channel and my Teams DMs since yesterday, then reply to Marco confirming I'll join the 3pm.

Job 4 — Documents & files

Reading several PDFs back-to-back to extract or compare is one of the most common chains on the fleet — and your AI can also create the output: a Word summary, an Excel tracker, a PowerPoint deck.

Read: pdf_read, word_read, excel_read, ppt_read; find files with finder_search, fs_search, onedrive_search_files
Act: word_create, word_append, excel_create, excel_write_cell, ppt_create; organize with onedrive_write_file, onedrive_move_file

Read these three vendor PDFs, build an Excel comparison of price, term, and SLA, and save it to the Procurement folder on OneDrive.

Job 5 — Look things up & track work

Beyond your apps, LMCP reaches live data and ticketing: look up market data with stocks_get_quote and stocks_get_chart, search and update ServiceNow incidents, or check your VPN with nordvpn_status. On-device image generation (Apple's Image Playground) is on the roadmap.

Install LMCP

Download LMCP and install it:

Open the downloaded .dmg from your Downloads folder
Drag Local MCP to your Applications folder
Open Local MCP from Applications — it appears in your menu bar

Takes about 30 seconds. No API keys, no OAuth. Your local AI clients are configured automatically.

Works With Every AI Client

One install, every assistant:

Local (auto-configured): Claude Desktop, Claude Code, Cursor, VS Code (Copilot / Cline), Windsurf, Zed
Web AIs via the encrypted Cloud Relay: ChatGPT, Claude.ai, Grok, Perplexity, and OpenClaw — a secure tunnel to your Mac; the work runs locally and nothing persists on a server

Frequently Asked Questions

Does my data leave my Mac?

No. LMCP runs the tools on your machine and reads directly from your installed apps — no cloud processing, no tokens, and nothing is stored on any server. The optional Cloud Relay is an encrypted tunnel that lets a web AI trigger a tool on your Mac; the result is produced locally and never persisted.

Which AI clients work with LMCP?

Any MCP-compatible client. Locally and auto-configured: Claude Desktop, Claude Code, Cursor, VS Code (Copilot / Cline), Windsurf, and Zed. Via the Cloud Relay: ChatGPT, Claude.ai, Grok, Perplexity, and OpenClaw.

Is LMCP really free?

Yes — free to use, no paid tier yet. Your license is permanent and never expires. If a paid plan is ever introduced it will be with advance notice, and early users keep free access. There is no lock-in.

The Best MCP Server for Mac Productivity in 2026