MCP Server + Flutter MCP Toolkit
@Arenukvern
MCP Toolkit for Flutter AI Agent Driven Development (MCP/CLI + custom client side tools) - via closed feedback loop (visual & semantic snapshot) and high client side customization adaptable for any Flutter app. Nowadays it is often called as agentic harness.
Overview
What is Flutter MCP Toolkit?
Flutter MCP Toolkit is a Dart MCP server and Flutter package that lets AI agents (Claude Code, Codex, Cursor, and others) inspect and drive a running Flutter app through semantic snapshots, widget interaction, hot-reload, log reading, and runtime custom tool registration.
How to use Flutter MCP Toolkit?
Install the binary via curl, add the toolkit to your Flutter app with flutter-mcp-toolkit codegen-init, install agent skills with flutter-mcp-toolkit init <agent>, then run your app with flutter run --debug.
Key features of Flutter MCP Toolkit
- 30 MCP tools under the
fmt_*capability prefix across four categories - Dynamic Tools Registration for apps to create custom tools at runtime
- Supports inspection, interaction, debug, and lifecycle operations
- Works with multiple AI agents including Claude Code, Codex, and Cursor
- Flutter apps can expose custom MCP tools and resources at runtime
Use cases of Flutter MCP Toolkit
- An AI assistant inspecting widget state, tapping buttons, and typing into forms in a running Flutter app
- A closed feedback loop where the AI hot-reloads code, verifies changes, and iterates without leaving the conversation
- A Flutter app registering its own domain-specific MCP tools at runtime for the AI agent to use
FAQ from Flutter MCP Toolkit
How does this differ from the official MCP Server for Flutter?
The official server exposes Dart tooling; this project focuses on dynamic MCP tools registration from within the Flutter app and a close feedback loop for AI agents.
What are the runtime requirements?
Flutter 3.44 or later. The Flutter app must be running in debug mode. The MCP server binary is shipped via GitHub Releases.
Are there any known limitations?
Dump RPC methods (e.g., dump_render_tree) can produce large token output and are disabled by default; enable them with the --dumps flag.
How is security handled?
The server connects to a Flutter app in debug mode only. It has been verified by MseeP.ai. Review the ARCHITECTURE.md for details on potential modifications for improved security.
Which AI agents are supported?
Claude Code, Codex, Cursor, Intent, Cline, and others. Use flutter-mcp-toolkit init <agent> or npx skills add Arenukvern/mcp_flutter -a <agent> to install skills.
