Misaki Studio supports two complementary ways to work: direct editing in the project interface and project operations through its production MCP server. You do not have to commit to one approach for an entire project.
you want to explore the file or element hierarchy;
visual placement, proportion, timing, or direct manipulation matters;
you are learning what controls are available for the current selection;
you want to compare responsive screens, variants, themes, or timeline states by eye;
you want to review and refine a change interactively.
The UI also contains session-only controls such as open tabs, Previous/Next file history, expanded rows, zoom, and panel sizes. These help you navigate, but they are not project content an agent needs to reproduce.
you can describe a bounded result more quickly than you can perform every step;
you want to inspect files, elements, styles, logic, timelines, or diagnostics in a structured form;
a repeated or multi-step change benefits from precise project, file, and node references;
you want a preview screenshot, DOM inspection, measurement, or other machine-readable verification;
you want to combine several low-level operations into a repeatable workflow.
Misaki Studio's production MCP server runs with the desktop application. A connected agent selects or addresses a project, then operates on project files and their content through registered tools.
Open the intended project in Misaki Studio.
Tell the agent the exact project, file, target element, desired result, and what it must not change.
Ask the agent to inspect the current state before editing.
Require the agent to read back or preview the result.
Review the affected file and element in the editor.
Refine manually, ask for a follow-up change, or undo the file-content mutation.
In the desktop editor, file tabs and File/Element Explorer rows can display MCP read and write activity. These indicators help explain which project areas an agent touched; they do not replace checking the resulting structure and preview.
The UI and MCP do not need identical gestures to reach an equivalent result.
Task: Read the project file tree · Coverage: UI + MCP · Important difference: The UI expands folders visually; MCP returns structured file data.
Task: Create, rename, move, duplicate, or delete supported files · Coverage: UI + MCP · Important difference: UI menus are capability-sensitive; MCP uses explicit references and schemas.
Task: Add or edit Component elements · Coverage: UI + MCP · Important difference: The UI uses the canvas, explorer, and inspector; MCP targets screen/node references.
Task: Edit styles, logic, or timelines · Coverage: UI + MCP, property-dependent · Important difference: Exact schema and availability must be checked for the specific operation.
Task: Reorder visible tabs or resize a panel · Coverage: UI only · Important difference: This changes the current editor session, not project content.
Task: Inspect preview DOM, computed styles, or interpreter state · Coverage: MCP only or MCP-led · Important difference: These operations are optimized for machine-readable inspection.
Task: Export to a local destination · Coverage: Desktop required · Important difference: The destination and export target still need exact workflow-specific verification.
Select the Website project. Read its file tree and summarize only its Pages and Components without changing anything. For each result, include the project-relative path. Stop after the read and wait for me to choose a file.
This prompt is safe because it names the project, limits the scope, requests structured evidence, and explicitly forbids mutation.
For a change, add a precise target and a proof step:
In the Website project, inspect the selected Component before editing. Add one Text element named
Labelinside its Body, set its text toContinue, then read the new node back and show me a preview screenshot. Do not change any existing element.
The exact tool sequence depends on the file and reference returned during inspection. A typical Component sequence uses project selection, element creation or text tools, and a preview read. The agent should not guess a node identifier that it has not inspected.
For setup and precise targeting, continue with What MCP means in Misaki Studio, Connect an agent, and Target files, folders, and elements.
The production MCP server exposes undo and redo for selected or referenced file-content history. These are different from the top bar's Previous and Next controls, which navigate file-selection history.
Whole-file operations and external actions such as deletion, export, Git, package publication, or website publication have different recovery models. Their guides must state those side effects separately; do not assume the selected file's content undo stack can reverse them.