Switchboard
Reads iMessage conversations from the local Messages database and sends approved replies through Messages.app, with rate limits and restrictions for new conversations.
Reads WhatsApp desktop conversations from its local database, supports searching and reading threads, and can open WhatsApp with a drafted reply typed in for the user to send.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Switchboard
Give your AI assistants your LinkedIn, iMessage and WhatsApp conversations, on your own Mac, with you approving every message they send.
Switchboard is a small Mac app that runs a local MCP server. Claude Code, Claude Desktop, OpenMausBot and other MCP clients can then search your conversations, find who's waiting on a reply, read a thread, and draft and send replies. You approve each reply on your Mac. Nothing goes to a server: your messages, your LinkedIn login and the index stay on your Mac.
Why
People who work with lots of outside contacts (founders, investors, recruiters, anyone in a "who's waiting on me?" job) end up with conversations spread across three apps. Their AI assistants can read their email and calendar but not these three apps, so arranging one meeting means copying text between LinkedIn, Messages, WhatsApp and a chat window.
Switchboard puts the three behind one set of tools, so you can say "who's waiting on a reply from me?" or "find my last thread with Ada and suggest a time for next week" and get an answer drawn from all three. It was built for one person's daily use and is shared in case it's useful to others.
Related MCP server: iMessage MCP Server
What it does
Reads LinkedIn messages (through allman), iMessage (from Messages' local database) and WhatsApp (from the WhatsApp desktop app's local database), plus Contacts to link one person across channels. It reads from temporary copies and never opens a live Apple or WhatsApp database.
Finds the threads that need a reply, grouped into 1:1, group, and unknown senders.
Sends only after you approve the exact text. LinkedIn sends through allman, and iMessage through Messages.app. For WhatsApp, the app opens with the text typed in and you press Enter.
Lets you choose, per channel, whether assistants can read nothing (Off), read only (Read only), or read and ask to send (Read & write). Switching a channel to Off deletes Switchboard's copy of it.
Install
You need an Apple Silicon Mac on macOS 14 or later. For LinkedIn you also need Google Chrome, Microsoft Edge, Brave or Chromium installed: the one-time sign-in window opens in it.
Download
Switchboard-<version>.dmgfrom the latest release.Open it and drag Switchboard into Applications, then open it. It lives in the menu bar and starts at login.
Follow the window:
Grant Full Disk Access, which it needs to read Messages, WhatsApp and Contacts.
Choose what each channel may do. Everything starts Off.
Sign in to LinkedIn if you turned LinkedIn on. allman opens a browser, and Switchboard never sees your password.
Connect your assistants.
More detail, including the command-line build, is in docs/INSTALL.md.
Connecting assistants
The Claude app (Claude Desktop) and Claude Code are connected for you each time Switchboard opens. Quit and reopen Claude afterwards so it loads Switchboard. The Claude app gets this entry in
~/Library/Application Support/Claude/claude_desktop_config.json, and the rest of that file is kept as it was:"switchboard": { "command": "/Applications/Switchboard.app/Contents/Helpers/switchboard", "args": ["mcp", "--client", "claude-desktop"] }OpenMausBot: the app's Connect Your Bots page copies a block to paste into Plugins → MCP servers.
Other MCP clients: run
switchboard mcp --client <label>as a stdio server, after adding<label>toclientsin~/Library/Application Support/Switchboard/config.json. Unknown labels are refused.
Tools
Read | Send |
|
|
Message bodies come back wrapped as { "untrusted_text": … }. Other people wrote them, so assistants are told to
treat them as data and never as instructions.
Safety and privacy
Local only. The only network traffic is allman talking to LinkedIn and your assistant talking to Switchboard. There is no telemetry.
Every send needs your approval of the exact text. You approve either in the assistant, when it can show a form, or on a Mac notification that opens a local approval page. An approval works once and expires after 10 minutes. If a new message arrives after you approved, the send is stopped. Nothing is retried automatically.
Rate limits. LinkedIn: 30 sends a day, 2 a minute, at least 30 seconds apart. iMessage: 30 a day. New iMessage conversations only go to people in your Contacts.
Logs hold no message content. The audit log records which tool was called, by which client, and the shapes and sizes of arguments.
Known gap: any program running as you on your Mac could reach the approval page, so an approval isn't proof that a human gave it. See docs/DESIGN.md.
About LinkedIn
LinkedIn has no public messaging API. allman uses the same private API as LinkedIn's website, with your own signed-in session. That may break when LinkedIn changes things, and you are responsible for following LinkedIn's User Agreement. Use it for your own conversations: not for bulk outreach, scraping or operating other people's accounts. See allman's responsible-use notes. Switchboard isn't affiliated with LinkedIn, Apple, WhatsApp or Meta.
Credits
LinkedIn support is built on allman by Tarka Ventures, a local-first LinkedIn messenger CLI. allman handles LinkedIn sign-in, sync and sending. Switchboard pins one allman release by checksum, bundles it, and never upgrades it on its own. allman is MIT-licensed. Its license and notice are in THIRD_PARTY_NOTICES.md and ship inside the app.
Also built with Bun, the MCP TypeScript SDK and SwiftUI.
Develop
bun install
./check # typecheck, lint (biome), tests
./scripts/fetch-allman.sh # the pinned allman, checksum-verified, into vendor/
./scripts/build-app.sh # dist/Switchboard.app, signed with a local dev identityTests use synthetic data only and never reach LinkedIn, Messages or WhatsApp. They run against fake stand-ins
for those (scripts/fake-*). docs/DESIGN.md explains the pieces,
docs/GLOSSARY.md the terms, and CLAUDE.md the rules for working in the repo.
Releases are in docs/RELEASING.md.
Issues and pull requests are welcome. Please keep real names, numbers and message text out of them.
License
MIT. Bundled third-party software keeps its own license (THIRD_PARTY_NOTICES.md).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Read and search your Instagram, WhatsApp and LinkedIn DMs and prepare follow-ups you approve.
Mac & Windows: let ChatGPT, Claude & Cursor use your email, calendar, iMessage, Teams, files.
iMessage for AI agents, from your iPhone. No Mac needed, and you choose which chats they read.
Your own WhatsApp in Claude and ChatGPT: read chats, draft replies, send messages you approve.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI assistants to read iMessage history and send messages on macOS. Supports conversation listing, message search with keyword and semantic modes, contact lookup, and sending messages to existing conversations.1311MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI assistants to read, search, and send iMessages, manage contacts, and access attachments on macOS.1MIT
- AlicenseAqualityBmaintenanceEnables AI agents on macOS to securely read and search the local Messages database, catch up on missed messages via a persistent inbox, and send texts or files to allowlisted chats, with optional voice note transcription and text-to-speech.1027 npmMIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents on macOS to search and summarize recent WhatsApp chats, draft replies, and send messages only after explicit confirmation, with scoped reads and constrained scheduled sends.2MIT