Artisan MCP Server
Provides tools for working with Dart projects: run the Artisan command registry, install and uninstall plugins through reversible declarative manifests, generate scaffolding and code (plugins, commands, CLI wrappers), and introspect the project through doctor/list surfaces. The server targets plain Dart CLI projects as well as Flutter apps.
Provides tools for driving and inspecting a running Flutter application from an AI agent: start/stop app sessions, hot reload and hot restart the running VM, stream logs, run preflight doctor checks, list available Artisan commands, and evaluate expressions against the live app through a tinker REPL. Also covers scaffolding and plugin installation for Flutter projects.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
At a glance
Fact | Value |
Package | |
Install |
|
Import |
|
Entry points |
|
Platforms | macOS and Linux for the lifecycle commands ( |
Requires | Dart >=3.13, Flutter for the app you drive |
Dependencies |
|
License | MIT |
Related MCP server: machine-mcp
Install
dart pub add dowel
dart run dowel installinstall writes bin/dispatcher.dart (the entry that calls runDowel), the codegen barrels lib/app/_plugins.g.dart and lib/app/commands/_index.g.dart, and builds the fast wrapper bin/dowel (about 0.2 s to start, against 0.4 to 2.7 s for dart run). Run it again and it skips every file that already exists (--force overwrites them) and keeps the dowel constraint your pubspec already has: install adds dowel: ^<running version> only when the pubspec declares none. It adds only the machine-local cache to .gitignore (.dowel/cli-bundle*, .dowel/build.stamp* and .dowel/.dowel.lock/), so commit .dowel/plugins.json and .dowel/installed/.
Moving an app from fluttersdk_artisan? install stops on the files that package generated and prints what to delete and the commands to run; doctor warns about the same leftovers. The steps are in Migrating from fluttersdk_artisan.
dart run dowel list # every registered command, grouped by namespace
dart run dowel doctor # preflight: Flutter and Dart on PATH, the default portRun the app under supervision
start launches flutter run detached and records the session per project under ~/.dowel/sessions/<hash>/, so two apps can be driven at once and a command never acts on another project's app.
dart run dowel start --device=chrome
dart run dowel reload # hot reload, Dart state kept
dart run dowel hot-restart # hot restart, Dart state dropped
dart run dowel logs --follow
dart run dowel stopCommand | Effect |
| Launch, tear down and relaunch the app; |
| Send |
| Print the recorded session as JSON; print or follow the captured log |
| A REPL (or |
A command that dials the running app (tinker among them) exits 75 while the app is still building, so a script can retry; any other non-zero code is a real failure.
Add a command
make:command Greet scaffolds the class and registers it in the nearest provider. The signature string declares the arguments and flags:
import 'package:dowel/dowel.dart';
final class GreetCommand extends DowelCommand {
@override
String get signature => 'greet {name} {--shout}';
@override
String get description => 'Greet a person by name.';
@override
CommandBoot get boot => CommandBoot.none;
@override
Future<int> handle(DowelContext context) async {
final name = context.input.argument('name') ?? 'world';
final greeting = 'Hello, $name';
context.output.success(
context.input.option('shout') == true ? greeting.toUpperCase() : greeting,
);
return 0;
}
}A command that the signature cannot express overrides configure(ArgParser) instead. Two providers registering the same command name fail at boot with DowelCommandCollisionException.
Install plugins
dart pub add awesome_plugin
dart run dowel plugin:install awesome_plugin --dry-run # preview every operation
dart run dowel plugin:install awesome_pluginA plugin describes its install footprint in an install.yaml manifest, or in code with the PluginInstaller DSL when it needs branching. Either way the installer stages every operation and writes nothing until the commit, writes through a .tmp file and a rename, skips what is already in place, and records each applied operation under .dowel/installed/<plugin>.json, so plugin:uninstall can reverse it. The reversal removes the files an install wrote, the dependencies it added and the plugin's registry entry; a dependency the project already declared stays, and edits injected into existing files (InjectImport, InjectAfterPattern) are logged as skipped, so remove those by hand. An install record written by an older dowel is refused until plugin:install <name> --force rewrites it, and plugin:install refuses to run when lib/app/_plugins.g.dart registers providers but .dowel/plugins.json is missing: restore it from version control.
Piece | Where it lives |
Manifest: files to copy, provider and config wiring, Android, iOS, macOS and web edits, env vars, prompts |
|
Procedural escape hatch: |
|
The plugin's commands | A |
A plugin without a manifest is registered directly in .dowel/plugins.json and needs the scaffold from dowel install. make:plugin <name> generates a plugin package with all of this wired.
Give your agent the MCP server
./bin/dowel mcp:install # or: dart run dowel mcp:installmcp:install writes the dowel entry into .mcp.json; reconnect your client once (Claude Code: /mcp reconnect dowel). The server speaks stdio JSON-RPC and exposes the substrate tools plus whatever the installed plugins contribute:
Tool | Does |
| Run the app lifecycle |
| Apply a code change to the running app |
| Read the session, the log (returned in the tool result), the toolchain and the command list |
| Evaluate a Dart expression in the running app; a blank |
dowel_start and dowel_restart take every option their CLI commands take. Without a running app the server stays up with the substrate tools and reconnects on the next call that needs the VM Service, after the app restarts too. A filter (.dowel/mcp.json, then DOWEL_MCP_* variables, then mcp:serve flags) allows or denies tools and packages; a deny wins at every layer.
Commands
Namespace | Commands |
Lifecycle |
|
Scaffolding |
|
Plugin management |
|
MCP |
|
Introspection |
|
Codegen |
|
For AI coding agents
Rules for an agent working with this package. Each one is a mistake the maintainers already made once.
Do | Because |
Run | The command registers into |
Preview with | File writes swap in through |
Commit | The registry regenerates |
Leave | They are generated, a hand edit is overwritten |
Start the app with | Reload, restart, logs and tinker find the app through the recorded session |
In a git worktree, run | The ownership check passes when the worktree sits inside the project root the main app recorded (an app at the repo root, the worktree under it) and for state with no recorded |
Reconnect the MCP client after | Clients do not reload a server entry on their own |
Use |
|
Read the exit code: | The app may still be building |
Never treat | It is this package's own entry point; the wrapper is |
Dependencies
dart_mcp for the MCP server, vm_service for the connection to the running app, args for command-line parsing, yaml and yaml_edit for manifests and in-place edits, xml for Android and Apple platform files, crypto for the install-record hashes. Dowel itself is pure Dart; the app you drive needs Flutter.
Links
Documentation: installation, quickstart, migrating from fluttersdk_artisan, every command, MCP setup, plugin authoring and the signature DSL
llms.txt and the agent skill for AI coding agents:
npx skills add anilcancakir/dowel
License
MIT, see LICENSE. Built by Anılcan Çakır.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Drive real devices from your AI Coding tool. Embed a client SDK (Unity, Godot, Flutter, iOS/macOS, Android, React Native, Web) in your app, then capture screenshots, traverse the UI tree, inject taps and key events, and run automated test tasks on the physical device over a secure relay.
Develop, manage, and debug Railway projects, services, and deployments from within agents.
Build, run, inspect, and continue AgentScript workflows on connected Super Chrome and Mac clients.
Flutter in-app bug tracking, visual pin annotations, and device logs directly into your AI IDE.
Related MCP Servers
- AlicenseCqualityDmaintenanceA distributable Model Context Protocol (MCP) server that exposes Dart SDK commands for AI-powered development. This server bridges the gap between AI coding assistants and Dart/Flutter development workflows by implementing the Model Context Protocol (MCP).106MIT
- AlicenseAqualityDmaintenanceEnables an AI agent to run local tools on the user's own machine via stdio, including command execution, workspace file read/write, and system status checks.52MIT
- AlicenseNot gradedqualityBmaintenanceExposes 35 stdio JSON-RPC tools that let AI agents snapshot the Flutter Semantics tree, drive gestures, capture screenshots, and observe live app state against running Flutter apps over VM Service.3MIT
- AlicenseAqualityBmaintenanceConnects AI coding clients to a running Flutter app's Dart VM Service so they can stream live exceptions with stack traces, logs, network calls, frame timings, memory, rebuild attribution and navigation state as structured tool results. It also correlates that evidence into root-cause diagnoses with confidence and suggested fixes, and streams everything to a browser dashboard.2244 npmMIT