widget_review_mcp 0.1.0

MCP server that lets AI agents read, reply to, and resolve Widget Review comments in a running Flutter app over the Dart VM Service.

widget_review_mcp

An MCP server (stdio) that lets AI agents read, reply to, and resolve the review comments a developer leaves by tapping widgets in a running Flutter app that uses widget_review.

Install

dart pub global activate widget_review_mcp

Register it with your agent, starting the agent from your Flutter project's directory:

claude mcp add widget-review -- widget_review_mcp

For Cursor, Windsurf, and other clients, add {"command": "widget_review_mcp"} to your mcp.json servers.

Finding the app

Run your app with:

flutter run --vmservice-out-file=.dart_tool/widget_review/vm_service_uri

If you start the app with the ▶ button, add that flag to your run configuration once: toolArgs in .vscode/launch.json for VS Code, or Additional run args in Android Studio / IntelliJ. See the widget_review setup for details.

The server reads that file (in the project directory or its subdirectories, newest first). You can also use --vm-service-uri, $WIDGET_REVIEW_VM_SERVICE_URI, or the connect_app tool with the URL flutter run prints. It reconnects on its own after hot restarts and new flutter runs.

Tools

get_reviews, get_review, add_comment, add_reply, resolve_review, unresolve_review, update_comment, delete_review, connect_app.

Each review includes the widget's source location and a snippet of the code around it, the visible text, accessibility info (label, role, tap-target size, likely issues), replies, and whether it was found on screen (match: exact | relocated | notFound).

Run widget_review_mcp --help for options.