Mac Messages MCP

A Python bridge for interacting with the macOS Messages app using MCP (Multiple Context Protocol).

<a href="https://glama.ai/mcp/servers/gxvaoc9znc"> <img width="380" height="200" src="https://glama.ai/mcp/servers/gxvaoc9znc/badge" /> </a>

Quick Install

For Cursor Users

Click the button above to automatically add Mac Messages MCP to Cursor

For Claude Desktop Users

See the Integration section below for setup instructions.

Features

• Universal Message Sending: Automatically sends via iMessage or SMS/RCS based on recipient availability
• Smart Fallback: Seamless fallback to SMS when iMessage is unavailable (perfect for Android users)
• Message Reading: Read recent messages from the macOS Messages app
• Contact Filtering: Filter messages by specific contacts or phone numbers
• Group Chat Filtering: Use chat IDs from tool_get_chats to read one group conversation chronologically
• Fuzzy Search: Search through message content with intelligent matching
• Attachments: Find and view photos, PDFs, and other attachments shared in conversations
• iMessage Detection: Check if recipients have iMessage before sending
• Cross-Platform: Works with both iPhone/Mac users (iMessage) and Android users (SMS/RCS)

Working with attachments

Attachment access uses progressive disclosure — discovery is cheap, fetching is deliberate:

1. Tier 1 — discovery in message search. tool_get_recent_messages and tool_fuzzy_search_messages annotate messages that have attachments with a compact summary like [attachments: #42 image/jpeg (invitation.jpg)]. The id lets you fetch the file later.
2. Tier 2 — attachment-first search. tool_search_attachments(start_date, end_date, contact, mime_type, limit) returns metadata only (id, MIME type, filename, size, sender) — useful for "find all images Elizabeth sent in April 2026" without scanning message text.
3. Tier 3 — fetch. tool_get_attachment(attachment_id) returns the file. Image MIME types come back inline (HEIC is converted to PNG so it can be viewed directly). PDFs, video, and audio come back as a filesystem path the agent can read with its own tools. Inline image bytes are capped at 5MB by default to avoid context blowup; oversized images fall back to path return.

Stickers, link-preview "balloon" payloads, and .pluginPayloadAttachment containers are filtered out by default.

Recipient formats

For direct sends, E.164 phone numbers with a leading + are the most reliable format, such as +14155551234. Bare digit phone numbers with a country code are normalized before sending, and 10-digit US numbers are sent as +1.... tool_find_contact returns phone matches in the same send-ready format.

Prerequisites

• macOS (tested on macOS 11+)
• Python 3.10+
• uv package manager

Installing uv

If you're on Mac, install uv using Homebrew:

brew install uv

Otherwise, follow the installation instructions on the uv website.

⚠️ Do not proceed before installing uv

Installation

Full Disk Access Permission

⚠️ This application requires Full Disk Access permission for your terminal or application to access the Messages database.

To grant Full Disk Access:

1. Open System Preferences/Settings > Security & Privacy/Privacy > Full Disk Access
2. Click the lock icon to make changes
3. Add your terminal app (Terminal, iTerm2, etc.) or Claude Desktop/Cursor to the list
4. Restart your terminal or application after granting permission

Integration

Claude Desktop Integration

Option 1: Claude Desktop Extension

This repo includes an MCPB-compatible manifest.json for Claude Desktop's one-click extension flow.

yarn global add @anthropic-ai/mcpb
mcpb pack

Install the generated .mcpb file from Claude Desktop Settings > Extensions > Advanced settings > Install Extension....

Claude Desktop, or the terminal used to package/run the extension, still needs Full Disk Access to read Messages.

Option 2: Manual Config

1. Go to Claude > Settings > Developer > Edit Config > claude_desktop_config.json
2. Add the following configuration:

{
    "mcpServers": {
        "messages": {
            "command": "uvx",
            "args": [
                "mac-messages-mcp"
            ]
        }
    }
}

Cursor Integration

Option 1: One-Click Install (Recommended)

Option 2: Manual Setup

Go to Cursor Settings > MCP and paste this as a command:

uvx mac-messages-mcp

⚠️ Only run one instance of the MCP server (either on Cursor or Claude Desktop), not both

Docker Container Integration

If you need to connect to mac-messages-mcp from a Docker container, you'll need to use the mcp-proxy package to bridge the stdio-based server to HTTP.

This repository also includes a Dockerfile for catalog checks and container builds:

docker build -t mac-messages-mcp .

Messages.app automation is macOS-only and will not work inside a Linux container. Container use is primarily for MCP catalog compatibility and read-only database experiments with mounted data.

Setup Instructions

1. Install mcp-proxy on your macOS host:

npm install -g mcp-proxy

1. Start the proxy server:

# Using the published version
npx mcp-proxy uvx mac-messages-mcp --port 8000 --host 0.0.0.0

# Or using local development (if you encounter issues)
npx mcp-proxy uv run python -m mac_messages_mcp.server --port 8000 --host 0.0.0.0

1. Connect from Docker: Your Docker container can now connect to:

• URL: http://host.docker.internal:8000/mcp (on macOS/Windows)
• URL: http://<host-ip>:8000/mcp (on Linux)

1. Docker Compose example:

version: '3.8'
services:
  your-app:
    image: your-image
    environment:
      MCP_MESSAGES_URL: "http://host.docker.internal:8000/mcp"
    extra_hosts:
      - "host.docker.internal:host-gateway"  # For Linux hosts

1. Running multiple MCP servers:

# Terminal 1 - Messages MCP on port 8001
npx mcp-proxy uvx mac-messages-mcp --port 8001 --host 0.0.0.0

# Terminal 2 - Another MCP server on port 8002
npx mcp-proxy uvx another-mcp-server --port 8002 --host 0.0.0.0

Note: Binding to 0.0.0.0 exposes the service to all network interfaces. In production, consider using more restrictive host bindings and adding authentication.

Option 1: Install from PyPI

uv pip install mac-messages-mcp

Option 2: Install from source

# Clone the repository
git clone https://github.com/carterlasalle/mac_messages_mcp.git
cd mac_messages_mcp

# Install dependencies
uv install -e .

Usage

Smart Message Delivery

Mac Messages MCP automatically handles message delivery across different platforms:

• iMessage Users (iPhone, iPad, Mac): Messages sent via iMessage
• Android Users: Messages automatically fall back to SMS/RCS
• Mixed Groups: Optimal delivery method chosen per recipient

# Send to iPhone user - uses iMessage
send_message("+1234567890", "Hey! This goes via iMessage")

# Send to Android user - automatically uses SMS
send_message("+1987654321", "Hey! This goes via SMS") 

# Check delivery method before sending
check_imessage_availability("+1234567890")  # Returns availability status

As a Module

from mac_messages_mcp import get_recent_messages, send_message

# Get recent messages
messages = get_recent_messages(hours=48)
print(messages)

# Send a message (automatically chooses iMessage or SMS)
result = send_message(recipient="+1234567890", message="Hello from Mac Messages MCP!")
print(result)  # Shows whether sent via iMessage or SMS

As a Command-Line Tool

# Run the MCP server directly
mac-messages-mcp

Development

Versioning

This project uses semantic versioning. See VERSIONING.md for details on how the versioning system works and how to release new versions.

To bump the version:

python scripts/bump_version.py [patch|minor|major]

Security Notes

This application accesses the Messages database directly, which contains personal communications. Please use it responsibly and ensure you have appropriate permissions.

License

MIT

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Star History