mcp-servers

MCP (Model Context Protocol) servers that give AI coding agents, GitHub Copilot in VS Code and Claude Code, the same inboxes and tools I use every day: Teams, Outlook, iMessage, WhatsApp, WeChat, Telegram, Discord, LinkedIn, Xiaohongshu, IMAP mail, a Markdown knowledge wiki and a job runner.

npm run demo starts brain-mcp against the made-up wiki in examples/sample-brain/ and calls two of its tools, the way an agent would. No accounts or API keys.

Write-up: mingz.dev/projects/mcp-servers

Why it exists

I run most of my day through agent chats: a morning scan of every channel, replies drafted in my voice, PR review asks, a task board. An agent is only as useful as what it can read. For the channels I actually use, servers either didn't exist (WeChat, iMessage with an SMS fallback, Xiaohongshu) or wanted broad tokens and sent messages on their own. These are small, run locally over stdio, and keep a person in front of anything that goes out.

Quick start

Needs Node 22 or newer (CI runs 22 and 24). This part needs no accounts:

git clone https://github.com/Mingz6/mcp-servers.git && cd mcp-servers
npm ci
npm run build   # every Node server, TypeScript strict
npm test        # brain-mcp, msgraph and outlook unit tests
npm run demo    # brain-mcp on examples/sample-brain, two tool calls over stdio

Python servers keep their own virtualenv:

cd packages/imessage    # or linkedin, telegram, wechat, xiaohongshu
python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
.venv/bin/python -m unittest test_mcp_server

Then register the servers you want with your client. Each one starts through its run.sh, which finds nvm's Node or the package's .venv:

claude mcp add brain -e BRAIN_MCP_ROOT=$HOME/notes -- $PWD/packages/brain-mcp/run.sh

// VS Code mcp.json
{
  "servers": {
    "brain": {
      "type": "stdio",
      "command": "${userHome}/code/mcp-servers/packages/brain-mcp/run.sh",
      "env": { "BRAIN_MCP_ROOT": "${userHome}/notes" }
    }
  }
}

Logins are per server and covered in each package's README: a device code for Microsoft, a QR scan for WhatsApp and Xiaohongshu, phone and OTP for Telegram, a one-time headed browser login for LinkedIn.

Servers

Server	Language	Talks to	What an agent can do	Status
msgraph	TypeScript	Microsoft Graph, work account	Teams chats, meetings and transcripts; Outlook mail with attachments; Loop pages; a one-way calendar copy to the Mac	daily use
outlook	TypeScript	Microsoft Graph, personal Outlook.com	read, search and send mail, attachments	daily use
imap-mail	TypeScript	IMAP (Gmail, QQ Mail, ...)	read, search, attachments; one instance per account	daily use
imessage	Python	macOS Messages (chat.db + AppleScript)	read, search, send; resends as SMS when Apple refuses an iMessage	daily use
whatsapp	JavaScript	WhatsApp Web via Baileys	read, search, send	daily use
wechat	Python	WeChat for Mac (SQLCipher database or screen OCR)	read, search, voice-note transcripts; read-only	daily use
telegram	Python	Telegram user API via Telethon	read, search, send	daily use
discord-user	TypeScript	Discord, as the user	read servers and DMs; read-only, never sends	daily use
linkedin	Python	linkedin.com via Patchright

Third-party servers (context7, Azure DevOps) run through scripts/npx-nvm.sh, which loads nvm's Node before npx.

Architecture

flowchart LR
  subgraph clients["Agent clients"]
    copilot["VS Code Copilot"]
    claude["Claude Code"]
  end

  subgraph servers["packages/, one process each"]
    graphServers["msgraph, outlook"]
    chatServers["imessage, whatsapp, telegram,<br/>wechat, discord-user, imap-mail"]
    webServers["linkedin, xiaohongshu"]
    localServers["brain-mcp, flywheel,<br/>worker-center-mcp"]
  end

  clients -- "stdio, started through run.sh" --> servers
  graphServers --> msalAuth["msgraph-auth<br/>MSAL device code + token cache"] --> graphApi[("Microsoft Graph")]
  chatServers --> chatStores[("chat.db, SQLCipher DB,<br/>Baileys, Telethon, Discord, IMAP")]
  webServers --> browserProfile[("persistent browser profile")]
  localServers --> localFiles[("Markdown wiki, SQLite")]

This is the beginning of the README. Read the rest at the original. Read the full README ↗