scene.read

Since 0.6Changed in next

[editor or disk] The scene as canonical JSON (argus://contract/scene): the workspace service's when one runs (an open editor's unsaved edits only once saved: dirty says they exist), else the editor's document (unsaved edits too), else the file.

Plain text

pointer: one value. id (* wildcard), tag, component: matching entities with index, a row each when many; pointer then reads each. summary: counts; a large scene's answer unless full.


Title	Read a scene
Mode
Token cost	about 252 tokens (in the default list, paid on every call)
Toolset	core

Arguments

Argument	Type	Required	Description
`agent`	`string`	no	The name of the agent making the call, when several agents share one connection: each named agent is its own actor and plays its own game. Leave it out when there is one agent.
`component`	`string`	no
`full`	`boolean`	no	`true` returns every finding.
`id`	`string`	no
`page`	`integer`	no	A page of the last findings, counted from 0, read without running the check again.
`pointer`	`string`	no	/entities/@door/transform: @x is the element whose id, else type, is x
`resolved`	`boolean`	no	Prefabs expanded, project defaults merged, sources marked
`scene`	`string`	no	The scene file to act on. The open scene when absent.
`summary`	`boolean`	no
`tag`	`string`	no

Example

An MCP tools/call with the required arguments (placeholders in angle brackets).

{
  "name": "scene.read",
  "arguments": {}
}