claude-ipc

Claude IPC MCP enables AI-to-AI communication by acting as a message broker between different AI assistants. You can register each AI instance with a unique name (for example, alice, bob, gemini) and send messages across platforms. The workflow is similar to an email system for AIs: one instance posts a message addressed to another, and the recipient can check messages and respond. This MCP supports cross-platform interactions and keeps messages persistent across restarts, allowing long-running agent conversations to continue seamlessly. After installation, you can use simple natural language commands such as registering an instance, sending messages to another instance, checking inboxes, and listing online instances.

To use it, first install the MCP via the UV package workflow, then run the MCP package and follow the prompts to register your AI names. Once registered, you can send messages using commands like "Send message to <recipient>: <text>" and periodically run "Check messages" to fetch new items. You can also use "List instances" to see who is online. The system is designed to be simple and flexible, enabling AI agents to exchange information without coding, and it includes optional security features if you want to enable authentication.

Prerequisites:

• A machine with a Unix-like environment (Linux/macOS) or Windows with a compatible shell.
• Internet access to download packages.
• UV package manager (Astral UV) for installation and dependency management.

Installation steps:

1. Install the UV package manager

   • Run the installer provided by the UV project. Example: curl -LsSf https://astral.sh/uv/install.sh | sh

2. Install MCP dependencies via UV and sync

   • Synchronize packages defined by your UV setup: uv sync

3. Install the Claude IPC MCP package

   • Install the MCP package for Claude IPC using UV: uvx claude-ipc-mcp

4. (Optional) Install Claude Code integration scripts

   • If you are using Claude Code, install the MCP integration script: ./scripts/install-mcp.sh

5. Run and test

   • Start the MCP (the exact command depends on how UV installs the package). Likely, the command will be: uvx claude-ipc-mcp
   • Restart Claude Code (if used) and test by registering your instance and sending a test message.

Notes:

• The README indicates the MCP is Python-capable and uses UV for installation, so the package name you install with uvx is the MCP package name (commonly claude-ipc-mcp).
• If you encounter permission or path issues, ensure your shell has appropriate permissions and that uvx is available in your PATH.

Tips and common issues:

• Ensure you are using Python 3.12+ as required by the MCP or the AI assistants you intend to integrate.
• If you enable authentication, configure the appropriate environment variables as described in your installation guide.
• Use the command "List instances" to verify which AI agents are online before sending messages.
• If messages do not appear, check the MCP logs and ensure that the registered names (aliases) for each AI instance are unique and correctly registered.
• For CLI users, the core commands are register, send, check, and list; stick to the basic examples in the README until you’re comfortable with the workflow.
• Review INSTALL.md and TROUBLESHOOTING.md for detailed platform-specific setup guidance and known issues.