LMCP Puts 190+ Mac App Tools Into Claude Code for Free

A free Mac app called LMCP exposes what its makers describe as 190-plus local tools to Claude Code, Cursor and Codex — most of them, they say, without connecting a single account.

The pitch targets a specific stall. Wiring MCP servers by hand means one server per app, an API key here, an OAuth registration there — and, as the LMCP team puts it, "for anything Microsoft, a trip to your IT admin for consent." Their argument is that the data is already on the machine, inside apps the user is logged into.

What the tools cover

Per the vendor's own description, a single install lets MCP clients work with Mail, Calendar, Teams, Slack, WhatsApp, iMessage, Outlook, OneDrive, Google Drive, Zoom, Office files, the browser and the terminal. The installer detects which AI apps are present and configures them; for Claude Code, the team says the tools are simply there at the next session start.

The local-cache trick

The mechanism worth understanding is where the data comes from. The MCP server runs on the Mac and the tools execute there, reading what apps already keep on disk rather than calling cloud APIs where possible.

Teams is the example the team volunteers: instead of Microsoft Graph, LMCP reads the local cache the Teams app keeps on disk, so there is no Azure app to register and no admin consent to chase. The same logic covers the browser tools, which the vendor says ride existing logged-in sessions, so no credentials change hands.

That design is the whole value proposition and also the thing to evaluate: it trades a reviewed OAuth scope for direct filesystem reads of application caches. Whether that is an improvement depends entirely on whose policy governs the machine.

Recipes and scheduled runs

Beyond one-off prompts, the vendor describes "recipes" — a multi-step workflow saved once and re-run on demand, their example being a morning pull of new orders from a dashboard into a sheet with anomalies flagged. Because Claude Code and Codex can run unattended, the team notes those workflows can be hooked into a script or a scheduled job.

What the vendor says it does not do

The limitations section is unusually direct for a launch post, and it is the part a buyer should read first. In the team's words: tools run on your Mac, "but your AI provider still sees whatever you ask it to read." LMCP does not make Claude or ChatGPT local. It does not bypass app permissions or company policies.

One architectural detail follows from that: web clients such as claude.ai or ChatGPT can reach LMCP only through an optional cloud connector, which ships off by default. Desktop and CLI clients do not need it.

Trying it

Installation is a shell one-liner, an npx command, or a signed installer:

curl -fsSL 'https://local-mcp.com/install?ref=devto-g1' | bash
# or
npx -y local-mcp@latest setup

Restart the AI client afterwards — MCP tools load at startup.

Two caveats before you pipe that first command to a shell. Every capability figure above is the vendor's own, published in a launch post, with no independent testing cited. And if the machine is company-managed, the local-cache approach is precisely the kind of access your security team will want to rule on before it reads a Teams cache on their behalf — the absence of an admin consent prompt is not the absence of an admin decision.