App Generation
Generating an App
Once your client is connected, describe the app to your agent the way you would describe it to a colleague:
Build me a tickets dashboard on my ToolJet DB.
The agent inspects the data sources available in your workspace, creates the app, adds the queries, binds components to them, and hands back links for editing and testing the finished app. Naming the data source and the entities you care about produces a much more useful first version than a generic request.
Your agent can talk to several MCP servers at once. If you also have Figma's MCP server connected, point the agent at a frame and let it read the design before it builds:
Read the Tickets dashboard frame in Figma, then build it as a ToolJet app on my ToolJet DB.
The agent pulls the layout, grouping, and copy from the design and maps them onto ToolJet's component library. Expect a close structural match rather than a pixel-perfect one, since the app is assembled from ToolJet components rather than from exported markup.
Modifying an Existing App
The same connection works on apps that already exist, whether they were built by hand, generated by ToolJet AI, or created earlier by your agent. Point the agent at the app by name and describe the change:
In my Tickets app, add a filter dropdown for priority above the table and wire it into the list query.
Agents with browser access can then open the app and verify the change actually works before handing it back.
An agent connected over MCP acts with the permissions of your access token and can run read queries against the data sources in that workspace. Because an agent's exploration is not deterministic and depends on the model you use, assume that any data reachable from that workspace may be read while it works. Scope the token to a workspace you are comfortable exposing.
By default, both ToolJet MCP and ToolJet AI operate only in the development environment, so an agent reads the data sources configured for development and not the ones behind staging or production. We recommend connecting MCP only to instances whose development environment points at development data.
The number of tokens consumed by an operation depends on the model you select, its reasoning effort setting, and the complexity of the request. These tokens are billed by your AI client's model provider, not by ToolJet.