Recipes
Once terradart-mcp is connected (Connecting clients), you mostly drive it through natural-language prompts — the agent picks the right tool. These recipes pair a goal with an example prompt and the tools the agent will reach for. See the Tools reference for exact inputs and outputs.
Resource discovery
Section titled “Resource discovery”When you want to know what TerraDart can provision for a given service, ask in plain language. The agent calls list_barrels to see the service surface, then list_resources (filtered by barrel) to enumerate the specific factories.
Prompt: “What does TerraDart support for Pub/Sub?”
The agent typically calls list_barrels (to confirm pubsub exists and how many resources it has), then list_resources with { "barrel": "pubsub" }, and summarizes the google_pubsub_* resources with their one-line descriptions.
Schema lookup
Section titled “Schema lookup”When you are writing a Stack and need the exact constructor for a resource, ask for it by name. The agent calls get_resource_schema to get the precise Dart class name, constructor parameters, nested types, and sensitive fields — so the code it writes actually compiles.
Prompt: “How do I construct a Cloud Run v2 service in TerraDart? Show me the constructor.”
The agent calls get_resource_schema with { "name": "google_cloud_run_v2_service" } and uses the returned constructor_signature and nested_types to write the GoogleCloudRunV2Service(...) call with the right parameters. If you give it a slightly wrong name, the tool’s suggestions field lets the agent self-correct.
Scaffolding
Section titled “Scaffolding”When you want a working starting point rather than a single resource, ask for a quickstart. The agent calls get_quickstart with a scenario key and drops the returned, CI-validated Stack into your project.
Prompt: “Scaffold a Cloud Run webhook stack with TerraDart.”
The agent calls get_quickstart with { "scenario": "cloud-run-webhook" } and uses the returned dart_code as the skeleton, then adapts names and regions to your needs. The four available scenarios are cloud-run-webhook, data-pipeline, lb-stack, and firestore-seeded-data; if the agent guesses a key that does not exist, the available list steers it to a valid one.
Architecture / plan review
Section titled “Architecture / plan review”When you already have a proposed stack (or a design sketch), use the catalog to validate it against real factory names before you write or run anything. The agent cross-checks each resource against list_resources and get_resource_schema, catching invented class names and wrong parameters early.
Prompt: “Here’s my planned TerraDart stack: a Pub/Sub topic, a push subscription, and a Cloud Run service that consumes it. Verify every resource and parameter against the TerraDart catalog before I implement it.”
The agent resolves each resource with get_resource_schema, flags anything that returns found: false (with suggestions), and confirms the constructor parameters exist — turning “looks plausible” into “compiles against the curated API.”