Loading ToolBench
Godot MCP has 50 tools with consistent naming (verb_noun pattern) and descriptions present for all. However, schema quality is uneven: many tools lack explicit input parameter type definitions in visible code, parameter descriptions are often generic, and output schemas are not documented. Error handling guidance is minimal. The server demonstrates good composition (tools are single-purpose) but falls short of production-grade clarity for LLM reasoning.
Apply patch-based changes to script files
Capture a screenshot of the editor or game viewport
Create new nodes in a scene
Create a new Godot project at the specified path
Create a new resource of a specified class
Create a new scene with a root node
Control debugger execution (pause, continue, step)
Delete assets from the project with preview capability
Output schemas not documented. Tools return results but LLMs cannot predict response structure, forcing them to guess at field names for downstream tool chaining.
Parameter descriptions are generic and lack format/constraint details. E.g., 'scope' in get_context is described as 'Context scope (e.g., progress)', unclear what valid values are or what 'progress' means.
Destructive operations (delete_assets, purge_deleted_assets, delete_nodes) lack dry-run or confirmation patterns. Agents cannot preview consequences before committing irreversible changes.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2025-06-18+ | v2 |
Delete nodes from a scene
Create or edit an animation
Edit an AnimationTree state machine or graph
Edit a tileset resource
Export and build the project for a platform
Find assets in the project with filtering and search options
Get animation data from an AnimationPlayer
Get information about a Godot class and its members
Get the current context and state of the Godot editor and project
Get script diagnostics and errors
Get export presets configured for the project
Get documentation and usage guides for the server and its tools
Get editor or runtime logs
Get the result of a pending operation with optional waiting
Get a resource and its properties
Get scene structure and properties
Get project settings
Get tilemap data from a TileMap layer
Import asset files into the project
Inspect the debugger state of a running game
Inspect properties of a running node
Install the Godot MCP plugin into an existing project
Move or rename assets in the project
Open a scene in the Godot editor
Paint tiles on a TileMap layer
Preview an animation at a specific time
Permanently purge previously deleted assets
Read script file contents
Restore previously deleted assets
Resume or reload script changes in the editor
Run a scene in the Godot editor
Sample performance metrics during runtime
Save scene and resource documents
Send input events to a running game
Set breakpoints in scripts for debugging
Stop a running game session
Undo a previous edit operation
Update properties of existing nodes
Update properties of a resource
Update project settings
Connect or disconnect signals between nodes
Wait for a runtime condition to be met
Error handling guidance is absent. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next (e.g., 'If operation_id not found, call get_context() first').
Parameter types are present in schema but descriptions do not explain constraints. E.g., 'recursive' is boolean but no description clarifies what recursion means in the search context.