App Canvas
Your whole MCP server on one screen, as a connected map.
App Canvas draws everything your server exposes as one map. The agent sits in the middle, and every tool, resource, and prompt it can use is connected to it. It's the fastest way to see the shape of a server: what it can do, what it reads, and what it asks the model to follow.
Read the map
Each card is one capability, labeled by type.
Tools show the tool's name, its description, and the names of its input fields, such as location, checkIn, and checkOut. Resources show the resource's name, its URI, such as booking://example or /widgets/hotel-cards, and its content type. A text/html resource is a widget. Prompts show the prompt's name and description.
The lines show what the agent can reach. A card with a weak or missing description stands out here; the travel planner above has two prompts still marked TODO: Add description. The model chooses what to call from those descriptions, so fix them before you test in chat.
Filter the view
The bar above the canvas counts every item and filters the map by type: All, Tools, Tasks, Resources, or Prompts. Filtering helps on a large server, where showing only the tools makes the map readable again.
Deploy from the canvas
When the project is linked to a NitroCloud app, the app's name appears at the top right, next to Deploy. Deploy sends the current build to that app; Deployments keeps the history. The refresh button reloads the server's capabilities after a change, if Watch is off.