> For the complete documentation index, see [llms.txt](https://help.intangible.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.intangible.ai/overview/intangible-mcp/what-your-agent-can-do.md). # What your agent can do The actions Intangible's Model Context Protocol (MCP) server exposes to an agent, by mode, in plain prose – what an agent can build, and where you finish by hand. Once an agent is connected to Intangible's Model Context Protocol (MCP) server, it can drive most of Studio. This page lists what an agent can do, grouped by the three modes, plus where you still finish by hand. This inventory reflects the current Beta, and capabilities are expanding. ## What an agent places in a scene: models, primitives, presets, populators An agent builds a scene from four kinds of things, the same ones you use by hand. Models are 3D assets pulled from the library by meaning – ask for "a mid-century armchair" and the agent searches and places the closest match. Primitives are cubes, spheres, planes, and other base shapes for blocking. Presets are complete spaces dropped in at once, like a furnished room. Populators are procedural generators for many similar objects, like a forest or a crowd, placed in one step. ## In Build mode, the agent constructs the set In Build mode, an agent creates a project and scene, then fills it. It searches the asset library semantically, places models and primitives, drops in presets and populators, extrudes shapes, adds and edits characters, and sets the environment – sun position, fog, terrain. It can read the current scene back before it adds more. > Build a small cafe interior: a counter along the back wall, four two-top tables, warm afternoon light through a window on the left. ## In Compose mode, the agent frames the shot In Compose mode, an agent moves the camera and creates framed shots. It positions the camera for a specific angle, sets up a shot, and animates objects and camera moves across the scene timeline. > Frame a low three-quarter shot of the counter with a 35mm look, then add a slow push-in. Ask for a set of shots and they arrive as named cards in the storyboard, so the board reads like a shot list rather than a pile of cameras. ![Compose mode after an agent framed a set of shots, each one a named card in the storyboard](https://1179478100-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FSWrjIqlgVAblx6vHGJiM%2Fuploads%2Fgit-blob-45c60cf34caa5d3014e14ca5e579bf91aca58cb7%2Fmcp-agent-shot-list-01.png?alt=media) ## In Visualize mode, the agent renders the image In Visualize mode, an agent renders an image from the shot you composed. It can build the prompt from the scene automatically, so the render reflects what is in the set rather than a guess from text. See [How the visualizer thinks](/overview/concepts/how-the-visualizer-thinks.md). > Render the shot, photoreal, golden hour. An agent can also set the aspect ratio and resolution on the render, so "make it 16:9 at 2K" is part of the same instruction rather than a trip back into Studio. ## How an agent paces a long job A scene of any size takes minutes, and an agent that disappeared for ten of them would be indistinguishable from one that had crashed. So the session has a rhythm worth knowing about. **It orients you before it starts.** On first contact the agent sets expectations: work in two windows, chat on one side and the Studio tab on the other, and the order is Build, then Compose, then Visualize. It will tell you that the viewport stays empty for the first stretch while it searches the asset library, and that objects land close rather than exact. **It stops at checkpoints.** On a large scene it places a representative piece, then pauses and asks you to look before it commits to the rest. A row of chairs gets built once and checked before the other two dozen go in. Correcting one row is cheap; correcting twenty-five is not. **It asks before rendering.** Rendering is the slow step and the one that spends credits, so an agent presents the framed shots and waits for you to choose which to render rather than rendering all of them. If you want everything, say so. **It tells you what it cannot do.** Some things are reachable by hand in Studio but not through the tools an agent has. When it hits one, it should say which, and hand you the two-drag version rather than burning tool calls on a workaround. ## Where you finish by hand An agent gets a scene about 80 percent of the way, then hands it back. Objects land close, not pixel-perfect. Fine positioning, the last nudge on framing, and small art-direction calls are faster to do yourself in Studio than to describe in another prompt. ## Writing a good brief for the agent An agent works best when the brief names the scene, the objects, the camera angle, and the render look. Vague briefs produce vague scenes. A brief that reads like a shot list – set, camera, render style – gets a usable scene in one pass. ## Related * [MCP (Model Context Protocol)](/overview/intangible-mcp.md) * [The three modes](/overview/concepts/the-three-modes.md) * [How the visualizer thinks](/overview/concepts/how-the-visualizer-thinks.md) * [Troubleshooting and FAQ](/overview/intangible-mcp/troubleshooting-and-faq.md) *Last updated 19 September 2026, against release 2.0.70. This inventory reflects the Beta MCP server and is expanding; verify against your current Studio build.* --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://help.intangible.ai/overview/intangible-mcp/what-your-agent-can-do.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.