An MCP server that makes Flutter/Dart codebases instantly understandable to AI agents. Instead of spending 60,000+ tokens exploring a project every session, the agent queries pre-analyzed context through 6 tools — reducing orientation costs by 90%. Built on the flutter_project_structure Dart package (pub.dev). Supports real-time file watching with --watch mode.
Overview
Prerequisites:
dart pub add dev:flutter_project_structure
Tools (6):
| Tool | Description |
|---|---|
get_architecture | Returns project type (app/package/plugin/monorepo), architectural layers, layer dependency graph, and entry points |
get_file_purpose | Returns the purpose (widget, model, service, bloc, etc.) and code metrics for any specific file |
get_dependencies | Returns detected naming conventions — file suffixes, class suffixes, and file purpose distribution |
get_conventions | Returns detected naming conventions — file suffixes, class suffixes, and file purpose distribution |
get_todos | Returns structured TODO/FIXME index with line numbers. Optionally filter by severity |
get_project_structure | Returns full project structure markdown — summary by default, or a specific section (tree, frameworks, architecture, metrics, etc.) |
Author: MSI Shamim License: MIT
Server Config
{
"mcpServers": {
"flutter-project-structure": {
"command": "dart",
"args": [
"run",
"flutter_project_structure",
"mcp-server",
"--watch"
]
}
}
}