Tools overview

Godot MCP Studio exposes a large set of MCP tools, grouped by area. The full, always-current catalog with per-tool risk badges lives on the Tools page. This page explains how they fit together.

How tools are organised

Tools are grouped into categories like project, files, scenes & scripts, scene editing, signals & groups, running the game, the running game itself, assets, animation, audio, input, tilemaps, 2D placement, 3D, UI layout, physics, theming & shaders, content editors (Theme, Curve, Gradient, TileSet, SpriteFrames, VisualShader), navigation, particles, project settings, debugging, editor scripting, class reference, and export & build.

Every client gets the whole tool set of its edition, with nothing to switch on. Lite has a focused part of the catalog and Pro has all of it; the Tools page marks which is which, and Editions & pricing explains the difference. Your AI can only call tools your permission mode allows.

Static vs live tools

• Static tools work on files on disk: reading and editing scripts, scenes, and resources. They work whether or not Godot is open.
• Live tools talk to the running Godot editor over a local, token-secured bridge. They need Godot open with the addon enabled. Examples: reading the live scene tree, editing the open scene, or inspecting the game while it runs.

If a live tool fails, it's almost always because the editor isn't open or the addon isn't enabled. See Troubleshooting.

The escape hatch

When the fixed tools don't cover something, run_editor_script runs an arbitrary GDScript snippet inside the editor with full access to the editor API. It's how an AI can automate bulk edits, generate content, or drive a one-off migration without you touching Godot. Because it executes code, Safe Edit asks you before each run; Full Control runs it without asking.

Risky actions

Destructive and command-executing tools ask you first in Safe Edit, with a sentence saying what is about to happen, and run without a prompt in Full Control. The question appears in the desktop app, or in your MCP client when the app is not running. The mode is the safety control: Safe Edit lets an agent build all day and still needs your yes for a delete or a run, and switching modes in the app applies to the very next tool call.

Image Studio tools

Pro agents can generate images and sprite animations with hosted Image Studio, or draw, edit, slice, and export art locally for free. A hosted request does not spend credits when the tool is called: the desktop app shows the exact quote and requires the activated device to sign the approval first. The agent asks once which route you want, unless you already said, and for game art it uses Image Studio instead of writing its own image scripts.

Image Studio checks each request from your AI as soon as it arrives and tells it exactly what to change, such as the camera angle, size or style. Requests that can't run never reach Approvals, and a waiting request that can no longer run is removed with the reason sent to your AI.

Generated frames can be handed to the existing SpriteFrames tools. Image tools do not create movement controllers, navigation, or gameplay scripts unless you ask for that as a separate Godot task.

The AGENTS.md advantage

When the app writes AGENTS.md / CLAUDE.md into your project, the model gets a compact overview of these tools (one line per category with the flagship tool names) and how to use them well. The complete catalog with every description is installed as the godot-mcp-tools skill under .claude/skills/ and .agents/skills/, with one reference file per category so the model reads only the area a task needs. Cursor, Continue, and Windsurf receive the same catalog as an on-demand godot-mcp-tool-catalog rule instead. It's optional but strongly recommended. See Connect your AI client.