Local, cross-platform MCP server (Node.js/TypeScript) that lets LLM agents interact with the user via OS notifications, CLI prompts and a persistent terminal chat.
https://github.com/ttommyth/interactive-mcpYour AI assistant is making assumptions again. It's writing code you didn't ask for, choosing configurations you never specified, and proceeding with half-understood requirements. Sound familiar?
The Interactive MCP Server fixes this by giving your AI assistant the ability to actually talk to you during its work—through OS notifications, CLI prompts, and persistent chat sessions.
Most AI workflows look like this: you give an instruction, the AI goes silent, then dumps a complete solution that might miss the mark entirely. No clarification, no confirmation, no human-in-the-loop feedback.
This creates friction in professional workflows where precision matters. You end up in endless revision cycles because the AI made reasonable assumptions that happened to be wrong.
Interactive MCP transforms your AI into a collaborative partner that can:
No more guessing games. No more starting over because the AI went down the wrong path.
Interactive Code Generation Your AI can ask "Should I use TypeScript or JavaScript?" instead of assuming. It can confirm architectural decisions, ask about naming conventions, and validate requirements before writing a single line.
Configuration Management When setting up development environments, the AI can prompt for specific versions, ask about optional dependencies, and confirm environment variables instead of using defaults that break your workflow.
Pair Programming Sessions Enable persistent chat sessions where you and your AI can discuss approaches, review code snippets, and iterate on solutions in real-time—all without leaving your terminal.
Long-Running Tasks Get desktop notifications when your AI finishes generating documentation, completes data processing, or needs input to continue a multi-step workflow.
request_user_input - Ask questions with optional predefined answersmessage_complete_notification - Send OS notifications for task completionstart_intensive_chat - Begin persistent terminal conversationsask_intensive_chat - Continue conversations within active sessionsstop_intensive_chat - End chat sessions cleanlyEach tool integrates seamlessly with Claude Desktop, VS Code, and other MCP clients.
Add this to your claude_desktop_config.json:
{
"mcpServers": {
"interactive": {
"command": "npx",
"args": ["-y", "interactive-mcp"]
}
}
}
That's it. Your AI can now interrupt its work to ask questions, confirm actions, and keep you informed.
Works on Windows, macOS, and Linux with native OS notifications and terminal integration. The server runs locally alongside your MCP client, maintaining privacy while enabling rich interaction patterns.
Configure timeouts, disable specific tools, or customize the experience through command-line flags. The tool adapts to your workflow, not the other way around.
Interactive MCP eliminates the biggest frustration in AI-assisted development: the communication gap. Your AI becomes a true collaborator that asks when it's unsure, confirms when it matters, and keeps you informed when it's working.
Stop debugging assumptions. Start building with confidence.