leancodepl/marionette_mcp

What it solves

Marionette MCP allows AI agents (such as Claude Code, Cursor, and Copilot) to interact with and test running Flutter applications. It bridges the gap between an AI's ability to write code and its ability to actually drive a live app to verify that the code works as intended, enabling automated AI-driven smoke testing and interaction.

How it works

It uses the Model Context Protocol (MCP) to connect an AI agent to a running Flutter app in debug mode. By adding a binding package (marionette_flutter) to the app, the agent can access a focused set of high-signal tools to inspect the widget tree, perform gestures (tap, swipe, scroll), enter text, and capture screenshots. It can also read logs and trigger hot reloads.

Who it’s for

Flutter developers who want to use AI agents to automate the testing, debugging, and verification of their mobile or desktop apps.

Highlights

• Runtime Interaction: Focuses on driving the live app rather than development-time code analysis.
• Focused Toolset: Provides a minimal set of actions (tapping, scrolling, text entry) to keep AI context sizes manageable.
• Custom Extensions: Allows developers to expose app-specific actions to the agent via custom extensions.
• Agent Skills: Includes a SKILL.md file that teaches AI agents how to set up and drive the app without needing the full README in context.
• Broad Compatibility: Supports multiple AI tools including Claude Code, Cursor, Gemini CLI, and GitHub Copilot.