# Flutter MCP Toolkit [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/Arenukvern/mcp_flutter  
**GitHub Stars:** 387  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/flutter-mcp-toolkit

## Description
Inspect and drive Flutter debug apps/games: semantic snapshots, search, and custom client tools

## Claude Desktop Quick Installation
Heuristic fallback — verify the package name and runner against the repository README before running it. Uses `npx` (confidence: low):

```json
"mcpServers": {
  "flutter-mcp-toolkit": {
    "command": "npx",
    "args": ["-y","flutter-mcp-toolkit"]
  }
}
```

## Documentation

## What Flutter MCP Toolkit MCP server does

Flutter MCP Toolkit MCP server gives an AI agent access to a running Flutter application through a Dart MCP server and a Flutter package. The agent can inspect the app’s semantic state, search the exposed interface, tap widgets, enter text into forms, trigger hot reload, and read application logs. These capabilities let an agent work against the current debug session rather than relying only on source code.

The app can also register custom MCP tools and resources at runtime through MCP Toolkit. This allows application-specific operations to be exposed alongside the built-in inspection and interaction capabilities. The README describes the result as a feedback loop: the agent observes the app, performs an action, reloads or checks output, and uses the resulting state as evidence.

## How it works

The Flutter MCP Toolkit MCP server runs alongside a Flutter app in debug mode. Its Flutter-side package provides the connection into the application, while the MCP server makes that connection available to an AI client. Agents can use the shared app state to inspect a semantic snapshot, interact with controls, and review logs without leaving the conversation.

The toolkit also presents a broader automation model based on observing, acting, and verifying. The same vocabulary is used across interactive MCP operations, the command-line interface, programmatic Dart scenarios through `flutter_mcp_harness`, and lifecycle-oriented runners such as `oka`. The provided material identifies `AutomationDriver` as the common contract for these automation paths.

## Setup and configuration

Install the command-line binary with the project’s installation script:

```bash
curl -fsSL https://raw.githubusercontent.com/Arenukvern/mcp_flutter/main/install.sh | bash
```

From a Flutter project, run `flutter-mcp-toolkit codegen-init`. This adds the `mcp_toolkit` dependency and emits a `main.dart` snippet for the app. Then install the agent skills for a supported client, for example:

```bash
flutter-mcp-toolkit init cursor
```

The README also lists `claude-code`, `codex`, `cline`, `agents-skills`, and `all` as initialization targets. Finally, launch the application with `flutter run --debug` and connect the configured agent. The repository advertises Flutter 3.44 or later.

## Tools and capabilities

The Flutter MCP Toolkit MCP server supports these workflows:

- Capture semantic snapshots of the running app.
- Search the app’s exposed interface.
- Tap widgets and enter text into forms.
- Hot-reload the Flutter application.
- Read application logs.
- Register application-specific MCP tools and resources at runtime.
- Use interactive `fmt_*` operations, CLI workflows, and Dart automation scenarios.

Exact available operations can depend on the app’s registered tools and the current toolkit version. The README also documents an action-invocation tier in which an app’s intent registry supplies actions and validates their arguments.

## Limitations and notes

The toolkit is intended for a running Flutter debug app, so it is not described as a general-purpose Flutter source analyzer or a replacement for a deployed-app testing service. The application must include the Flutter package and be started in debug mode before the agent can inspect or drive it.

Custom capabilities are application-dependent: an app must register its own MCP tools or resources before an agent can use them. Client setup is also separate from app setup; installing the binary and adding the package does not by itself configure every AI client. The README names Cursor, Codex, Zed, Claude Code, and Cline among the supported agent environments.

_Full upstream README: https://allmcps.com/mcp/flutter-mcp-toolkit/readme_