For local development, use the MCP Inspector.
Run the inspector with no arguments:
npx @modelcontextprotocol/inspector
Open the MCP Inspector in the browser and enter dart_mcp_server in the “Command” field.
Click “Connect” to connect to the server and debug using the MCP Inspector.
The server entrypoint lives at bin/main.dart, and can be ran however you choose, but the easiest way is to run it as a globally activated package.
You can globally activate it from path for local development:
dart pub global activate -s path .
Or from git:
dart pub global activate -s git https://github.com/dart-lang/ai.git \ --git-path pkgs/dart_mcp_server/
And then, assuming the pub cache bin dir is on your PATH, the dart_mcp_server command will run it, and recompile as necessary.
Note:: For some clients, depending on how they launch the MCP server and how tolerant they are, you may need to compile it to exe to avoid extra output on stdout:
dart compile exe bin/main.dart
And then provide the path to the executable instead of using the globally activated dart_mcp_server command.
After compiling the binary, you can run the example workflow client to interact with the server. Note that the workflow bot sets the current directory as the root directory, so if your server expects a certain root directory you will want to run the command below from there (and alter the paths as necessary). For example, you may want to run this command from the directory of the app you wish to test the server against.
dart pub add "dart_mcp_examples:{git: {url: https://github.com/dart-lang/ai.git, path: mcp_examples}}" dart run dart_mcp_examples:workflow_client --server dart_mcp_server
Modify your mcp.json file to run your locally compiled server instead of using the server from the Dart SDK.
Each time you make changes to the server, you'll need to restart the server on the MCP configuration page or reload the Cursor window (Developer: Reload Window from the Command Palette) to see the changes.