Use the SciStudio GUI

SciStudio supplies the running GUI address. Your AI client supplies the browser or computer-use tools that open, inspect, and operate it. Use the tools actually available in the current session and follow their browser/app selection and interaction instructions; do not assume a provider-specific tool name.

Connect to the running workspace

1. Call mcp__scistudio__open_gui. Open its complete returned URL with your browser or computer-use tool on the same machine as SciStudio. Preserve any deployment prefix and project/workflow query parameters: url attaches to the current project view; base_url omits that context. open_gui returns an address; it does not open a browser or click anything.
2. Reuse a browser tab already connected to that instance and intended project when available. Computer use can also operate the existing SciStudio desktop window. Do not guess a port or start a second backend.
3. Read the current page or take a screenshot. Confirm the intended project and content before acting. A new browser tab may have different selected items, open tabs, and active workflow from the desktop window.

If no suitable tool is available, or it cannot reach the instance, explain the specific limitation. Having a URL alone does not establish GUI access.

When you run inside an AI app through WebMCP

When the user runs SciStudio in External AI mode, the workspace is already open in your AI app's built-in browser, beside the conversation, and your SciStudio tools come from that page. Operate that page with the browser or side-panel tools and methods your own app provides for its built-in browser; they see and act on the same view the user sees, which is smoother than any other route.

• Work in the page that is already open. Do not open the address in a separate browser or tab: a page outside the AI app is not connected to this conversation, and the user will not see your actions there.
• The layout differs from the desktop window: the workspace sidebar is on the right, Preview sits in its own card, and the AI Chat tab is hidden because the conversation happens in your app.
• screenshot_gui is not available over WebMCP; use your app's own page screenshot or snapshot to observe the result.

Find the right part of SciStudio

Use visible labels, tooltips, and the current page structure. The sidebar may be on either side, so navigate by names rather than fixed screen positions.

The workspace sidebar has six sections: Blocks, Workflows, Data types, Data, MiniApps, and Project; a Preview section appears when there is something to preview. The bottom panel has the tabs AI Chat, Terminal, Config, Logs, Plots, History, and Git.

Goal	Where to look and what to do
Open a workflow	Open Workflows, find the intended workflow, and open its canvas tab.
Find a project file	Open Project, expand its folders, and open the file.
Find available blocks	Open Blocks. Inspect the block's name and description before selecting or dragging it onto the canvas.
See which data types exist	Open Data types.
Change a node's settings	Select the intended canvas node, then edit its parameters in the Config tab. Confirm the node identity when labels repeat.
Inspect data	Open Data, or select the relevant input/output port on the canvas to show its preview.
Open or manage a MiniApp	Open MiniApps.
See execution progress or errors	Check canvas node states, then the Logs tab.
See past runs and where a result came from	Open the History tab.
See or restore versions, or switch branches	Open the Git tab.
See a plot	Open the Plots tab.
Switch open content	Select the intended centre tab by its title; do not assume the most recently opened tab is still active.

Clicking the active workspace-section icon collapses its sidebar content; clicking it again opens it. Hover an unfamiliar icon to read its tooltip. For more detailed feature navigation, read user-guide/using-the-gui.md in the project. Prefer the observed interface if its layout differs from an example.

Operate controls

• With browser tools, locate controls using supported page snapshots, accessible names, roles, or visible text. With screenshot-based computer use, derive click and drag positions from the current image and refresh it after layout changes.
• Focus the intended field before typing. For dropdowns, inspect the options after opening them. For sliders and canvas dragging, observe both the control and the destination/result rather than assuming the gesture succeeded.
• Handle the visible dialog or menu before trying to click behind it. Use its labelled actions and read validation messages when an action is unavailable. Native file pickers may require computer use even when the page is controlled through browser tools.
• Embedded content may appear inside a frame. Use frame-aware browser operations when supported, or the visible view through computer use; a control absent from a top-level page snapshot is not necessarily absent from the interface.

Observe the result

For a screenshot of the existing SciStudio desktop workspace, local MCP may provide screenshot_gui(target="workspace"). For a visible MiniApp tab it also accepts target="miniapp" and an optional panel_id. It captures images only; use browser or computer-use tools for actions. It does not capture ordinary browser tabs, and the text-only WebMCP bridge does not support its image response. Use your browser tool's screenshot capability when operating a browser tab.

After an action, inspect the resulting page state or screenshot. Check the visible effect relevant to the task: the intended tab opened, the field changed, the data view updated, or the dialog completed. Wait for the relevant loading state to settle before judging it or issuing the action again.

If something fails, read the visible error and use available console or runtime logs to understand it. Avoid repeated clicks on an unchanged failure. Continue from the last observed state and report any unresolved limitation accurately. An opened URL is not proof of a rendered view, and a screenshot alone is not proof that an interaction worked.

Available tools

The live MCP tool list is the source of truth. SciStudio supplies only the tools below; clicking, typing, and page snapshots come from your own browser, side-panel, or computer-use tools.

Tool	What it does	When to use it
open_gui	Returns the address of the running GUI.	First, before opening SciStudio in a browser (not needed when you already run inside its page through WebMCP).
screenshot_gui	Captures the SciStudio desktop workspace or a visible MiniApp as an image (local MCP only).	To see the desktop window's current state.
list_panels	Lists existing panels; kind="miniapp" gives the MiniApps and their panel_ids.	To find the MiniApp to open.
open_miniapp	Asks the open workspace to open a MiniApp tab on a block output.	To show the user a MiniApp on their data instead of telling them where to click.
get_active_workflow_context	Returns the workflow open in the GUI.	To confirm which workflow the user is looking at.