twake-mcp

MCP server for Twake Workplace — connect any AI assistant to Linagora's collaboration suite.

What it does

Exposes Twake Workplace (Chat, Mail, Drive) as MCP tools that AI assistants like Claude, GPT, or Linagora's own LUCIE can use to interact with your workspace.

An AI assistant with twake-mcp can:

• Read and send chat messages
• Check your email inbox and read individual emails
• Browse, read, and upload files to Drive
• Search across all services simultaneously

Tools

Tool	Description
twake_chat_send	Send a message to a chat room
twake_chat_rooms	List joined rooms
twake_chat_history	Get recent messages
twake_mail_inbox	List inbox emails
twake_mail_read	Read a specific email
twake_mail_mailboxes	List mail folders
twake_drive_list	List files and folders
twake_drive_read	Read a file
twake_drive_upload	Upload content
twake_search	Search across all services

Setup

npm install

Configure via environment variables:

# Twake Chat (Matrix)
export TWAKE_MATRIX_HOMESERVER=https://matrix.twake.app
export TWAKE_MATRIX_TOKEN=syt_...
export TWAKE_MATRIX_USER=@jacob:twake.app

# Twake Mail (JMAP)
export TWAKE_JMAP_URL=https://jmap.twake.app/jmap
export TWAKE_JMAP_TOKEN=eyJ...

# Twake Drive (Cozy)
export TWAKE_COZY_URL=https://jacob.twake.app
export TWAKE_COZY_TOKEN=eyJ...

Usage with Claude Code

Add to your Claude Code MCP settings:

{
  "mcpServers": {
    "twake": {
      "command": "node",
      "args": ["/path/to/twake-mcp/src/index.js"],
      "env": {
        "TWAKE_MATRIX_HOMESERVER": "https://matrix.twake.app",
        "TWAKE_MATRIX_TOKEN": "your-token",
        "TWAKE_JMAP_URL": "https://jmap.twake.app/jmap",
        "TWAKE_JMAP_TOKEN": "your-token",
        "TWAKE_COZY_URL": "https://jacob.twake.app",
        "TWAKE_COZY_TOKEN": "your-token"
      }
    }
  }
}

Then ask Claude: "Check my Twake inbox" or "Send a message to #engineering on Twake"

Architecture

twake-mcp/
└── src/
    └── index.js          # Single-file MCP server
                           ├── API clients (Matrix, JMAP, Cozy)
                           ├── Tool implementations (10 tools)
                           └── MCP server setup (stdio transport)

┌──────────────────┐       ┌──────────────┐       ┌──────────────┐
│   AI Assistant   │       │   twake-mcp  │       │    Twake      │
│ (Claude, LUCIE,  │──────▶│  MCP Server  │──────▶│  Workplace    │
│  GPT, etc.)      │ stdio │              │ HTTPS │  (Chat/Mail/  │
│                  │◀──────│              │◀──────│   Drive)      │
└──────────────────┘       └──────────────┘       └──────────────┘

The server uses the @modelcontextprotocol/sdk stdio transport. Each tool maps directly to a Twake service API call — Matrix Client-Server API for chat, JMAP (RFC 8620/8621) for mail, and Cozy API for drive.

Zero dependencies beyond the MCP SDK. All HTTP calls use Node.js native fetch.

Why

Linagora is building LUCIE, an open-source multilingual LLM. This MCP server makes Twake Workplace AI-accessible, enabling LUCIE (or any AI) to interact with the collaboration suite. It's the bridge between Linagora's AI strategy and their product ecosystem.

License

AGPL-3.0 (matching Linagora's licensing)